WEBVTT - The Race to Regulate AI Has Begun 0:00:00.640 --> 0:00:03.960 Well, it's exciting al having actually Rishi Sunac do this 0:00:04.040 --> 0:00:08.239 global AI summit to try and shape the agenda on 0:00:08.280 --> 0:00:10.160 how you regulator and how you use it safely. 0:00:10.280 --> 0:00:13.200 I think people that are really grappling for the extent 0:00:13.280 --> 0:00:15.760 to which this is positive and the extent to which 0:00:15.800 --> 0:00:20.799 the truly alarming headlines around the damage it could also 0:00:20.920 --> 0:00:24.240 do the extent to which they are something real going concerns. 0:00:24.600 --> 0:00:26.320 And it's interesting to see the UK. Actually, I think 0:00:26.320 --> 0:00:28.800 it's the first summit of its kind, right, global summit 0:00:28.840 --> 0:00:31.280 looking at the risks of AI, and it's interesting to 0:00:31.280 --> 0:00:35.239 see the UK trying to gather everyone together to talk 0:00:35.240 --> 0:00:37.880 about it. 0:00:37.880 --> 0:00:41.720 It is interesting, and I think it feels convincingly Rishi Sunac, 0:00:41.760 --> 0:00:45.000 doesn't it. It feels like something everything about his backstory 0:00:45.240 --> 0:00:49.440 suggests he has a kind of credible take on AI, 0:00:50.320 --> 0:00:53.160 having worked in the States where many of these tech 0:00:53.159 --> 0:00:57.440 companies are based, and also at the same time wanting 0:00:57.480 --> 0:01:00.800 to just rein in and check some of its influence 0:01:00.920 --> 0:01:03.720 for you know, everyday voters and people. So I think 0:01:03.760 --> 0:01:07.880 there is It does feel like it's convincingly Sunakian. Not 0:01:08.360 --> 0:01:11.279 a word, maybe a word. I don't think it is, well, 0:01:11.720 --> 0:01:14.600 just any think it feels like he you know, but 0:01:14.680 --> 0:01:16.800 we we you know a lot of people are asking 0:01:16.880 --> 0:01:19.680 what actually are the aims for the summit, what actually 0:01:19.760 --> 0:01:21.160 do they want to come out of it. We've heard 0:01:21.200 --> 0:01:24.679 talk of this institute being set up. That's probably the 0:01:24.680 --> 0:01:27.440 easy bit. What is the institute do is the heart 0:01:27.480 --> 0:01:27.880 of it. 0:01:27.760 --> 0:01:32.840 And that's really what we're looking at this week. Welcome 0:01:32.840 --> 0:01:35.160 to In the City, Bloomberg's podcast, connecting you to the 0:01:35.160 --> 0:01:38.080 conversations and the story is shaping the world of finance. 0:01:38.319 --> 0:01:41.280 I'm front Sine Lackwell with Rastratton in the London studio. 0:01:41.600 --> 0:01:44.520 Now with us this week. Nick Langson, his global editorial 0:01:44.600 --> 0:01:47.840 lead for Tech and Gaming and co host of the 0:01:47.840 --> 0:01:51.640 Bloomberg original program AI I R L. In Real Life, 0:01:52.280 --> 0:01:55.360 Nate is the pace the fastest we've ever seen it 0:01:55.360 --> 0:01:56.360 at the pace for change. 0:01:56.600 --> 0:01:59.320 It feels like it, absolutely, and I think that is 0:01:59.360 --> 0:02:02.400 symptomatic the fact that the amount of money that now 0:02:02.440 --> 0:02:05.640 exists in order to invest in some of these technologies 0:02:05.920 --> 0:02:10.040 has never been greater. We have seen an enormous speeding 0:02:10.120 --> 0:02:12.760 up of tech development. Absolutely because a lot of the 0:02:12.800 --> 0:02:15.600 companies that are able to invest enormously in these kinds 0:02:15.639 --> 0:02:18.480 of technologies in artificial intelligence are the ones that have 0:02:18.560 --> 0:02:21.920 made the most money from the most recent technological innovations, 0:02:22.000 --> 0:02:24.120 ie cloud computing, search, and social media. 0:02:25.040 --> 0:02:28.040 Some of us are freaked out about AI. Others think 0:02:28.160 --> 0:02:29.920 that this is the best thing that could happen to us. 0:02:30.000 --> 0:02:31.960 But because of your time, why do you've seen these 0:02:32.040 --> 0:02:35.000 changes and really have chronicled them from the very beginning. 0:02:35.480 --> 0:02:38.480 I remember when DeepMind was founded and everyone got very 0:02:38.480 --> 0:02:41.679 excited about this weirdly secretive company that no one knew 0:02:41.680 --> 0:02:44.880 anything about. That suddenly was all the rage in computer 0:02:44.960 --> 0:02:47.800 science and deep learning, which is what underpins the vast 0:02:47.800 --> 0:02:51.600 majority of what we currently see as artificial intelligence, although 0:02:51.639 --> 0:02:55.280 that is only a fraction of the overall body of 0:02:55.360 --> 0:02:57.520 work that is AI, which has been in development since 0:02:57.520 --> 0:03:00.520 the nineteen fifties, and we've seen it go from a 0:03:00.520 --> 0:03:03.840 conversation about making better use of big data. You know, 0:03:03.880 --> 0:03:06.160 that was a huge topic about ten years ago. But 0:03:06.400 --> 0:03:09.040 obviously it goes without saying almost that the release of 0:03:09.080 --> 0:03:11.720 chat GPT, which bear in mind, at the time we 0:03:11.760 --> 0:03:15.000 speak right now isn't even a year old. I think 0:03:15.040 --> 0:03:17.480 that's sometimes quite shocking to people just to realize how 0:03:17.520 --> 0:03:21.240 relatively recent that all happened. It wasn't until that release 0:03:21.840 --> 0:03:24.680 was made public that people and businesses and investors and 0:03:24.720 --> 0:03:29.200 governments and regulators started to realize, Okay, now we get it. 0:03:29.320 --> 0:03:34.240 Now we can see the potential impact returns on investment 0:03:34.360 --> 0:03:38.040 but also harms. And it really, I think motivated everybody 0:03:38.080 --> 0:03:40.600 to think, Okay, now we need to really take this serious. 0:03:40.640 --> 0:03:43.440 And Nate haven't covered this for eighteen years when the 0:03:43.560 --> 0:03:46.240 last few weeks feels like the headlines and the warnings 0:03:46.400 --> 0:03:49.120 sort of end of the world is nigh have reached 0:03:49.120 --> 0:03:51.120 a bit of a kind of fever pitch. Do you 0:03:51.200 --> 0:03:54.360 think it's hysterical or do you think these are proportionate warnings. 0:03:54.520 --> 0:03:59.920 I think they are well meaning but wildly doom focus, 0:04:00.280 --> 0:04:03.440 and I think history teaches us some lessons here about 0:04:03.520 --> 0:04:06.640 how to prepare for these kinds of changes. My favorite 0:04:06.680 --> 0:04:08.800 comparison to make is that of the dot com boom. 0:04:09.120 --> 0:04:11.840 You know, in the very late nineties, there was an 0:04:11.960 --> 0:04:15.960 enormous fear that the millennium bug, as it was coined, 0:04:16.200 --> 0:04:20.600 was going to cause a complete collapse of society, planes 0:04:20.640 --> 0:04:23.480 falling out of the sky. There was at least one 0:04:23.480 --> 0:04:25.640 airline that put one of its senior managers on a 0:04:25.720 --> 0:04:28.479 plane just to prove that they weren't worried their planes 0:04:28.480 --> 0:04:30.039 were going to fall out of the sky at midnight. 0:04:30.360 --> 0:04:33.640 Banks printed additional cash because they feared people would do 0:04:33.680 --> 0:04:36.320 a run on the banks and stockpile cash. There was 0:04:36.320 --> 0:04:39.080 a legitimate fear that this was the end of all times. 0:04:39.120 --> 0:04:42.640 And then of course the millennium came to pass and 0:04:42.680 --> 0:04:46.120 everyone realized that maybe they'd slightly overplayed the worries. But 0:04:46.320 --> 0:04:48.960 it did encourage people for those years leading up to 0:04:49.000 --> 0:04:52.360 that to at least check and prepare. And that's why 0:04:52.440 --> 0:04:55.599 I think that the questions about is this going to 0:04:55.680 --> 0:04:59.200 usher in an era of homegrown chemical warfare? Will this 0:04:59.279 --> 0:05:02.279 cause the color of society? Because an AI becomes self 0:05:02.680 --> 0:05:06.800 interested in preserving itself and growing itself. And it is 0:05:06.880 --> 0:05:08.760 useful to ask those questions, but I think it's more 0:05:08.880 --> 0:05:11.640 useful to keep it into context around what is physically 0:05:11.680 --> 0:05:14.480 possible and what are some of the actual real world 0:05:14.520 --> 0:05:16.080 requirements for that to even happen. 0:05:16.160 --> 0:05:16.440 I e. 0:05:16.760 --> 0:05:19.800 An enormous amount of power and computing resource that either 0:05:19.839 --> 0:05:23.520 doesn't exist or exists but is fundamentally inoperable because it 0:05:23.520 --> 0:05:24.800 would just cost too much money. 0:05:25.160 --> 0:05:28.880 And you can see Richie Sunak and his government trying 0:05:28.880 --> 0:05:33.160 to straddle that sort of looking both ways. Huge risks coming, 0:05:33.279 --> 0:05:35.400 but huge opportunities too. Yeah. 0:05:35.440 --> 0:05:39.360 Absolutely, I mean it's a totally realistic concern and it's 0:05:39.400 --> 0:05:41.479 smart to think about these. It's very smart to have 0:05:41.520 --> 0:05:44.720 a public debate. And one of the biggest challenges that 0:05:44.760 --> 0:05:47.359 we have with social media now is the result of 0:05:48.320 --> 0:05:55.520 largely unchecked, unregulated growth of technology companies that eventually resulted 0:05:55.560 --> 0:05:57.960 in the consolidation of power and influence in the hands 0:05:57.960 --> 0:06:01.560 of a very small number of extraor ordinarily powerful entities, 0:06:01.839 --> 0:06:06.000 all of which are commercially motivated and publicly owned. And therefore, 0:06:06.240 --> 0:06:08.279 you know, we can't get the genie back in the 0:06:08.279 --> 0:06:10.240 bottle for that. We want to avoid that this time. 0:06:10.320 --> 0:06:12.440 I mean, that's my question when I'm looking ahead to 0:06:12.480 --> 0:06:14.680 the summit that they're hosting, which is in many ways, 0:06:14.720 --> 0:06:17.200 you know, is smart, and I think it probably is 0:06:17.240 --> 0:06:20.000 a decent moment for the UK to show some leadership 0:06:20.160 --> 0:06:22.320 in an area where there is great investment in the UK. 0:06:22.680 --> 0:06:26.080 But the question I keep returning to is, as you say, 0:06:26.160 --> 0:06:28.320 when you look at the evidence of getting big social 0:06:28.400 --> 0:06:30.880 media companies to do the right thing and to police 0:06:30.920 --> 0:06:34.240 themselves and so on, have we got the kind of 0:06:34.279 --> 0:06:36.919 track record from them that suggests that they will also 0:06:37.000 --> 0:06:40.839 police themselves in managing these huge language sets. 0:06:40.960 --> 0:06:44.640 Absolutely not no, there is absolutely no, they're absolutely not. 0:06:44.880 --> 0:06:47.360 But there are good intentions. But some of the good 0:06:47.360 --> 0:06:50.119 intentions also come with the caveat that they are self 0:06:50.200 --> 0:06:53.120 interested in protecting their business. They want to release the 0:06:53.160 --> 0:06:56.240 technology because they want to commercialize it and monetize it 0:06:56.279 --> 0:06:57.839 and be seen, as you know, as a leader in 0:06:57.880 --> 0:07:00.480 that field. But at the same time, you don't want 0:07:00.480 --> 0:07:02.040 to be caught out. And it's why you see a 0:07:02.040 --> 0:07:05.080 lot of companies calling for regulation, because it gives them 0:07:05.360 --> 0:07:07.760 boundaries and guardrails and it allows them to say, well, look, 0:07:07.760 --> 0:07:09.280 we've played by the rules you set, so it's not 0:07:09.279 --> 0:07:11.880 our fault that this happened. And that is a challenge 0:07:11.920 --> 0:07:12.520 in its own right. 0:07:12.600 --> 0:07:15.000 But it'd be interesting to see what you think actually 0:07:15.640 --> 0:07:19.200 the overarching goal for this AI summit is, and when 0:07:19.240 --> 0:07:22.920 we understand the goal, whether other countries like chrying to 0:07:23.040 --> 0:07:24.080 share those goals. 0:07:24.680 --> 0:07:28.840 Well, so yes to different different questions, but you're absolutely 0:07:28.880 --> 0:07:32.920 right the key questions what's the goal for the UK? Well, 0:07:33.280 --> 0:07:36.160 there is investment in the UK by these firms. If 0:07:36.160 --> 0:07:38.440 we're asking the question that we are often asking on 0:07:38.440 --> 0:07:41.200 this pod and in the readout and all over Bloomberg. 0:07:41.240 --> 0:07:43.640 Whereas the growth agenda come from, Partly it will be 0:07:43.680 --> 0:07:46.800 around investment in tech firms into the UK. So the 0:07:46.800 --> 0:07:49.320 first thing, there's that. Secondly it's something which she soon 0:07:49.360 --> 0:07:52.280 that cares about. And everything about his backstory points to 0:07:52.320 --> 0:07:55.880 a summit on AI. You know, if his ties right, 0:07:56.160 --> 0:07:59.000 you know, he's a you know, Standford Business School. He 0:07:59.040 --> 0:08:01.440 has spend a lot of his career in America, so 0:08:01.520 --> 0:08:05.520 he is instinctively comfortable with those companies. But equally he 0:08:05.720 --> 0:08:08.520 is the MP for constituency in Yorkshire, which you know, 0:08:08.880 --> 0:08:11.720 at the end of the day, normal people will be 0:08:11.760 --> 0:08:14.240 affected by this revolution that is coming. I think it 0:08:14.320 --> 0:08:16.600 is interesting the question around what effect it will have 0:08:16.640 --> 0:08:19.400 on jobs Eventually. We've had warnings this week on job 0:08:19.440 --> 0:08:21.880 losses by twenty thirty and are people ready? And you 0:08:21.920 --> 0:08:24.240 have a lot of very sensible people saying in that 0:08:24.320 --> 0:08:27.480 context where many many low paid jobs are being automated, 0:08:27.480 --> 0:08:30.520 we have to think about a debate around universal basic income, 0:08:30.560 --> 0:08:33.719 which a few years ago was unthinkable. I think we 0:08:34.120 --> 0:08:36.360 all know and everyone listening though, is that there are 0:08:36.440 --> 0:08:39.120 and Nate's mentioned it already there's these constant fars that 0:08:39.280 --> 0:08:41.080 new tech's going to get rid of jobs and they 0:08:41.120 --> 0:08:44.520 don't necessarily materialize. I feel a bit more gloomy about 0:08:44.559 --> 0:08:47.040 this one because you have seen increasingly over the last 0:08:47.080 --> 0:08:49.520 few years that lots of low paid jobs are being 0:08:49.559 --> 0:08:54.120 automated and they aren't being replaced by particularly high esteemed 0:08:54.600 --> 0:08:57.280 valuable jobs that give people, you know, a sense of 0:08:57.280 --> 0:08:58.760 purpose in the morning when they go to work. I 0:08:58.760 --> 0:09:00.480 think there is a question mark there, But it is 0:09:00.520 --> 0:09:03.520 the case too that history does show us that actually 0:09:03.559 --> 0:09:07.319 it's the kind of the luod Oitte fallacy that essentially, 0:09:07.679 --> 0:09:10.280 in the end, different jobs are made and there's more 0:09:10.320 --> 0:09:11.560 of them. Well, I hope. 0:09:11.600 --> 0:09:16.080 So is there illegitimacy for the UK to be hosting this? 0:09:16.200 --> 0:09:19.240 So by regulating it, do you also attract growth and 0:09:19.360 --> 0:09:21.680 investment because you're seen as the AI destination? 0:09:22.440 --> 0:09:24.760 I think we need to work very hard to be 0:09:24.800 --> 0:09:28.559 seen as an AI destination. I think that in the sense, 0:09:28.600 --> 0:09:31.199 the UK has an opportunity here to play something of 0:09:31.240 --> 0:09:35.680 a Switzerland role in AI because let's be honest, the 0:09:35.720 --> 0:09:38.840 money is in China and the US we do not 0:09:39.040 --> 0:09:43.000 have a huge AI tech scene in that you know, 0:09:43.080 --> 0:09:46.160 we are investing a tiny amount of money in AI. 0:09:46.240 --> 0:09:48.760 We are making a very small amount of money available 0:09:48.760 --> 0:09:52.320 to AI development. Even the billion pounds that figure that 0:09:52.360 --> 0:09:55.560 gets thrown around that you know, SUNA can smaller than 0:09:56.280 --> 0:09:59.040 I mean, I think it was Amazon invested something like 0:09:59.120 --> 0:10:04.120 four billion dollars in one AI company. That goes four 0:10:04.160 --> 0:10:07.600 times the figure from one company in one startup, like 0:10:07.920 --> 0:10:11.160 it pales into its significance. But I do think on 0:10:11.360 --> 0:10:14.280 my Switzerland comparison is that there is an argument I think, 0:10:14.280 --> 0:10:16.120 and it's not one the government I'm sure we'd agree with, 0:10:16.160 --> 0:10:18.400 but I think it's realistic that if you have a 0:10:18.400 --> 0:10:22.200 country like the UK that will be impacted or benefited 0:10:22.240 --> 0:10:26.000 by negative or positive outcomes of AI but doesn't actually 0:10:26.040 --> 0:10:29.560 have much of a commercial stake in its development, it 0:10:29.559 --> 0:10:33.520 actually puts it in a rather arguably neutral role in 0:10:33.600 --> 0:10:36.760 setting at least some of the agendas that other countries 0:10:36.880 --> 0:10:39.720 play by, because we can be a little bit more 0:10:39.760 --> 0:10:41.080 impartial as long. 0:10:41.040 --> 0:10:44.600 As whatever the UK's goals are shared with the people 0:10:44.600 --> 0:10:45.840 that matter, i e. China. 0:10:46.280 --> 0:10:49.120 Yes, that's very true, and there is some argument whether 0:10:49.160 --> 0:10:51.960 we do or not, And certainly the US and China 0:10:52.040 --> 0:10:54.520 have a very different approach to how they foster and 0:10:54.600 --> 0:10:57.240 encourage the growth and development of a lot of technologies. 0:10:57.320 --> 0:10:59.760 China is very much top down, the party will dictate 0:10:59.760 --> 0:11:01.600 a lot of the time where the money should go 0:11:01.679 --> 0:11:04.040 and where the support for development should come from, whereas 0:11:04.040 --> 0:11:06.040 the US is much more like, we'll get on with it, 0:11:06.120 --> 0:11:07.959 private enterprise and we'll regulate if we have to. 0:11:08.400 --> 0:11:11.760 And the pitch from the government will be, Okay, let's 0:11:11.840 --> 0:11:15.280 use these brexit freedoms. You've seen already tech bosses saying 0:11:15.480 --> 0:11:17.439 use your Brexit freedoms, right, So it's not just the 0:11:17.600 --> 0:11:21.120 necessarily a sort of comm's line from this government, it 0:11:21.200 --> 0:11:23.560 is it's also being echoed by people in the industry. 0:11:23.640 --> 0:11:26.640 So that's so that's that's that's the pitch. But completely 0:11:26.880 --> 0:11:29.280 you know, Nate is obviously very lucid on you know, 0:11:29.360 --> 0:11:33.280 in reality, does the UK have the heft to convene 0:11:33.320 --> 0:11:37.240 this debate and necessarily pull out of it's that kind 0:11:37.240 --> 0:11:40.640 of not quite as regulatory as the EU, not quite 0:11:40.679 --> 0:11:44.520 as wild west as America, and definitely not China or 0:11:44.559 --> 0:11:47.240 whether actually in all of that, you know, then you 0:11:47.280 --> 0:11:48.320 know they get left behind. 0:11:48.760 --> 0:11:50.720 But are they still aligned with the EU? So again 0:11:50.760 --> 0:11:52.000 it's I. 0:11:51.960 --> 0:11:54.640 Think there is that here that here that they can 0:11:54.720 --> 0:11:57.800 actually push out further and that and they're being encouraged 0:11:57.840 --> 0:11:59.720 to by various players in this scene. 0:12:00.120 --> 0:12:03.040 So they're forming an institute to try and understand. First, 0:12:03.080 --> 0:12:05.600 you have to understand it, I guess, right, So that's 0:12:05.640 --> 0:12:09.880 where we die. Where's that incident going to be? But 0:12:09.920 --> 0:12:12.160 you have to understand what you're regulating before you regulate. 0:12:12.200 --> 0:12:14.319 So the problem with regulators is that they're always a 0:12:14.320 --> 0:12:18.559 step behind. Because AI evolves, the chips evolve, it gets smarter, 0:12:18.679 --> 0:12:20.840 and the regulators need to catch up. 0:12:21.000 --> 0:12:22.960 That's always been the problem with technology as well. And 0:12:22.960 --> 0:12:24.679 that's what's so interesting is that none of this is 0:12:24.800 --> 0:12:27.800 unique to AI. This is something that's happened every year, 0:12:27.880 --> 0:12:31.000 every decade, seemingly that I can remember reading about or 0:12:31.040 --> 0:12:35.440 talking about. The hope is that we collectively learn from 0:12:35.440 --> 0:12:38.320 past experience that when we spot something that maybe we 0:12:38.320 --> 0:12:40.200 should probably put a lot of extra time into looking 0:12:40.280 --> 0:12:43.400 at early, we do that. I like to think that 0:12:43.400 --> 0:12:45.520 that's kind of what the world is doing with this, 0:12:46.600 --> 0:12:48.040 and that maybe that's a good thing. 0:12:48.760 --> 0:12:50.880 So Nate, isn't the problem with deep fakes that actually 0:12:50.960 --> 0:12:54.000 you can't really tell them apart. So there's a picture 0:12:54.040 --> 0:12:57.120 of a leopard going around. There's three of them, and 0:12:57.360 --> 0:12:59.680 it challenges you to find the deep fake one. 0:13:00.120 --> 0:13:03.640 The difficult it's very, very difficult. There are many experts 0:13:03.800 --> 0:13:06.319 trained in this that cannot tell a deep fake from 0:13:06.520 --> 0:13:09.280 the real thing, and so there's this debate around we 0:13:09.320 --> 0:13:12.360 should have everybody in the industry that is producing something 0:13:12.400 --> 0:13:14.360 that would qualify as a deep fake or something that 0:13:14.400 --> 0:13:18.400 is generated to water market. And that's not watermarking in 0:13:18.440 --> 0:13:21.840 the sense of if you look at a photo online 0:13:21.840 --> 0:13:26.080 and it's got like the brand slightly proof, yeah, proof 0:13:26.480 --> 0:13:29.520 across the top, or it's some audio water marking that 0:13:29.600 --> 0:13:31.800 makes the video useless to share or sell. It's not 0:13:31.800 --> 0:13:35.400 that sort of thing. It's baked into the image itself. 0:13:35.440 --> 0:13:38.480 In a very similar way to right now, if you 0:13:38.679 --> 0:13:41.520 take your iPhone or your Android phone, you take a photograph, 0:13:41.679 --> 0:13:44.040 the amount of data that is stored in that image 0:13:44.040 --> 0:13:46.760 that you never see unless you look for it, it 0:13:46.800 --> 0:13:50.080 is staggering. Everything is in there where it was taken, 0:13:50.559 --> 0:13:53.920 the GPS coordinates, the timestamp, the model of camera, the 0:13:53.920 --> 0:13:57.439 model of lens in the camera, the aperture of the 0:13:57.760 --> 0:14:01.280 lens attached to the camera. You know whether the flash fired, 0:14:01.320 --> 0:14:04.400 whether it didn't the model make everything. This would just 0:14:04.440 --> 0:14:08.480 be another piece exactly. It's all stored in. There is removable, 0:14:08.559 --> 0:14:12.480 you can delete it, but the vast majority are never 0:14:12.679 --> 0:14:14.480 going to be removed because no one knows they're there, 0:14:14.640 --> 0:14:15.760 and why would you. 0:14:15.840 --> 0:14:20.480 For the most that feels in a very difficult debate. 0:14:20.560 --> 0:14:24.040 That feels promising, But the problem is does everybody agree 0:14:24.080 --> 0:14:24.680 to the watermark. 0:14:24.720 --> 0:14:26.720 Everyone would have to agree to it, and everyone would 0:14:26.720 --> 0:14:28.440 have to sign up to support it in a way, 0:14:28.680 --> 0:14:31.080 and it's a slightly strange comparison perhaps, but in the 0:14:31.080 --> 0:14:33.440 way that if you buy an audio like an AV 0:14:33.640 --> 0:14:35.640 receiver for a home cinema or a pair of headphones 0:14:35.720 --> 0:14:38.040 or something that you know they all support Dolby Digital, 0:14:38.400 --> 0:14:40.120 you know that is a format, that is a standard, 0:14:40.200 --> 0:14:42.640 that is a thing that is internationally recognized and adopted, 0:14:43.000 --> 0:14:46.360 and so you would need a standard's body to say 0:14:46.400 --> 0:14:49.520 this is the standard for watermarking whether an image or 0:14:49.560 --> 0:14:51.680 even part of an image is a deep fake, and 0:14:51.720 --> 0:14:54.080 that every other piece of software or hardware around the 0:14:54.080 --> 0:14:57.480 world can read and interpret that, and that at least 0:14:57.520 --> 0:15:00.200 gets you some of the way towards identifying whether is 0:15:00.240 --> 0:15:02.560 real or fake. But it does change because you can 0:15:02.600 --> 0:15:04.040 manipulate images. 0:15:03.720 --> 0:15:06.200 And how quickly can that happen, because clearly we've got 0:15:06.200 --> 0:15:09.600 two major elections next year years so not for next year. 0:15:09.760 --> 0:15:13.400 Very quickly it can be developed, but adoption and an 0:15:13.440 --> 0:15:17.360 acceptance internationally and roll out is is a many years 0:15:17.360 --> 0:15:18.440 long process, I would say. 0:15:18.520 --> 0:15:20.800 So we'll see deep fakes in the next elections, right. 0:15:22.680 --> 0:15:24.560 I spoke to an expert the other day and said, 0:15:24.600 --> 0:15:28.360 will will deep fakes decide the twenty twenty four election, 0:15:28.720 --> 0:15:30.840 and he said, no, it'll be one or lost on 0:15:30.880 --> 0:15:34.000 abortion rights, but it will be deep fakes that are 0:15:34.200 --> 0:15:36.800 the core driver underneath that or words to that effect. 0:15:36.800 --> 0:15:38.600 It is by the time you've established that something's a 0:15:38.600 --> 0:15:40.360 deep fake, some of the damage will have been done 0:15:40.360 --> 0:15:42.160 twelve as later, and a lot of people won't even 0:15:42.160 --> 0:15:44.440 listen to the clarification, as we've seen recently. 0:15:44.560 --> 0:15:46.960 Yeah, and the argument comes down to, well, is being 0:15:47.040 --> 0:15:50.200 used to spread disinformation and sometimes things are created with 0:15:50.360 --> 0:15:56.640 the goal of subverting reality as a disinformation item. But 0:15:56.800 --> 0:15:59.840 if that looks real and somebody unwittingly shares it, they 0:15:59.880 --> 0:16:03.680 are sharing it effectively is misinformation because they don't realize 0:16:03.680 --> 0:16:06.160 that it's not real. And with the speed of social media, 0:16:06.760 --> 0:16:11.920 it can travel way quicker than you could ever possibly counteract. 0:16:11.440 --> 0:16:13.160 It, and so I mean, we're probably guilty in this 0:16:13.160 --> 0:16:17.040 conversation of focusing too much on the downside, and that's 0:16:17.080 --> 0:16:21.280 for justifiable reasons, but the upsides are massive. My favorite is, 0:16:21.280 --> 0:16:23.840 you know, for a teacher who has to mark thirty 0:16:23.840 --> 0:16:26.760 books or twenty five books every night, that can be aied. 0:16:28.360 --> 0:16:32.440 Ditto for difficult healthcare. AI will spot stuff that it 0:16:32.520 --> 0:16:35.840 was impossible for GPS to necessarily crawl over all that data. 0:16:35.920 --> 0:16:40.200 So there's legitimate hopes that sensible, proportionate use of AI 0:16:40.760 --> 0:16:44.240 can help with productivity and release all of us to 0:16:44.280 --> 0:16:46.880 do what The question is, how do you adjudicate well? 0:16:46.920 --> 0:16:48.880 I think one thing that's really important to keep in 0:16:48.880 --> 0:16:51.480 mind in all of these conversations is that there is 0:16:51.800 --> 0:16:54.720 this sort of common belief that you can take a 0:16:54.760 --> 0:16:58.000 problem and you can throw AI at it and that 0:16:58.080 --> 0:17:02.720 problem gets solved. And that's just not true. And it's 0:17:02.800 --> 0:17:05.880 really important that people understand that you can't just throw 0:17:05.960 --> 0:17:09.000 something into AI and suddenly get something better. It's a 0:17:09.040 --> 0:17:11.959 significantly more complex series of steps to do that. One 0:17:12.000 --> 0:17:13.960 of the things I find most fascinating about what could 0:17:13.960 --> 0:17:17.000 happen now is if you look at the large language models, 0:17:17.000 --> 0:17:20.000 which is what if you take chat GPT, that is, 0:17:20.440 --> 0:17:22.280 you know, the poster child for a lot of this, 0:17:22.600 --> 0:17:25.840 even though the technology underneath it isn't that new. It 0:17:25.960 --> 0:17:28.960 uses large language models, it scrapes the entire Internet, it 0:17:29.000 --> 0:17:32.280 reads every word on the Internet. Increasingly, now those sorts 0:17:32.280 --> 0:17:34.080 of products are able to look at every video and 0:17:34.119 --> 0:17:39.040 every picture and that's what's called a multimodal AI and 0:17:39.480 --> 0:17:42.320 it can process that, train itself on that and then 0:17:42.359 --> 0:17:44.840 give you usable outcomes. But there comes a point I think, 0:17:44.920 --> 0:17:47.480 and I actually think we're reaching that point now where 0:17:47.560 --> 0:17:50.760 everything that exists has been read and scanned. So in 0:17:50.840 --> 0:17:54.400 order to improve to the same degree, you need new 0:17:54.440 --> 0:17:57.800 types of data, and you need quality data. 0:17:57.960 --> 0:18:01.119 Reliable data, I mean reliable problem exactly, deep fakes right, 0:18:01.160 --> 0:18:03.000 it has to be vetted data. 0:18:03.040 --> 0:18:06.040 It does. And so the next step is, well, where 0:18:06.040 --> 0:18:08.920 does the next huge volume of data come from. It 0:18:08.960 --> 0:18:11.879 will come from sensors, it will come from us, it 0:18:11.880 --> 0:18:16.040 will come from medical trials and studies. But my point 0:18:16.119 --> 0:18:18.720 is is that if you develop some kind of a 0:18:18.800 --> 0:18:24.040 system that can try with molecules, looking at the interactions 0:18:24.280 --> 0:18:27.240 of molecules in a way that you cannot realistically do. 0:18:27.280 --> 0:18:30.679 In a lab, you can run all these simulations to 0:18:30.760 --> 0:18:34.080 figure out how a molecule might interact, take the data 0:18:34.160 --> 0:18:38.720 that that simulation has created, and then apply AI to well, 0:18:38.800 --> 0:18:42.000 what could we do with this that changes the game entirely. 0:18:42.080 --> 0:18:44.239 That is not where we're at really right now, but 0:18:44.280 --> 0:18:46.560 that is probably one of the next steps that we're 0:18:46.560 --> 0:18:48.560 going to get to, and that's where things get very exciting. 0:18:48.680 --> 0:18:53.600 An optimistic, I'm optimistic, but I mean I'm an optimist 0:18:53.640 --> 0:18:57.280 generally speaking, and I think that we have a when 0:18:57.280 --> 0:18:59.879 we talk about doomerism and AI doomerism, I think we 0:19:00.080 --> 0:19:03.080 take a really anthropocentric view of these systems in that 0:19:03.160 --> 0:19:07.400 we build things to kind of mimic our own inherent 0:19:07.480 --> 0:19:10.760 abilities with the human brain. Neural networks are very loosely 0:19:10.800 --> 0:19:13.480 modeled on how human neurons work and interact with each other, 0:19:13.840 --> 0:19:16.119 and so when they do something, as in, when an 0:19:16.160 --> 0:19:19.480 AI system does something that looks slightly human like, it 0:19:19.520 --> 0:19:22.040 looks a bit like creativity, It's kind of like seeing 0:19:22.040 --> 0:19:24.760 a face in a piece of toast, you sort of think, well, 0:19:24.840 --> 0:19:27.240 is that the you know, is that the reincarnation of 0:19:27.520 --> 0:19:31.680 the Messiah. It's not it's just very burn bread. But 0:19:32.359 --> 0:19:35.040 as human beings, we are evolutionary trained to see that, 0:19:35.119 --> 0:19:37.000 and I think we spot those in AI as well, 0:19:37.160 --> 0:19:39.040 and then we say, well, if it's going to do 0:19:39.080 --> 0:19:42.600 a little bit of human like creativity on this, humans 0:19:42.680 --> 0:19:45.720 do sadly have a tendency to start wars and kill 0:19:45.720 --> 0:19:48.360 each other, and maybe we should stop it doing that. 0:19:48.400 --> 0:19:50.520 But it's not necessarily coercentric. 0:19:50.760 --> 0:19:52.919 That's a big word name. I think I have to 0:19:52.960 --> 0:19:54.119 google it it. 0:19:54.480 --> 0:19:57.440 I mean, yeah, we obsession about human beings, and. 0:19:57.400 --> 0:20:00.919 It's the obsession that humans are the end game, you know. 0:20:00.960 --> 0:20:04.480 It's this thing that we are the most evolved and 0:20:04.560 --> 0:20:07.960 the holy Grail, the final step in the evolutionary journey, 0:20:08.119 --> 0:20:10.720