WEBVTT - Why AI Isn't Actually Boosting Productivity 0:00:02.720 --> 0:00:07.200 Bloomberg Audio Studios, Podcasts, radio News. 0:00:17.560 --> 0:00:20.759 I'm Stephanie Flanders, head of Government and Economics at Bloomberg, 0:00:21.000 --> 0:00:24.120 and this is Trumpnomics, the podcast that looks at everything 0:00:24.320 --> 0:00:28.639 in the economic world of Donald Trump. Today. I wanted 0:00:28.680 --> 0:00:33.160 to dig into two seemingly conflicting stories about the global 0:00:33.200 --> 0:00:36.280 economy that are often embedded in the conversations on this show. 0:00:36.360 --> 0:00:39.280 I mean, one story that we hear often is that 0:00:39.320 --> 0:00:43.040 we live in an era of extraordinary innovation, new more 0:00:43.120 --> 0:00:47.280 powerful AI models unveiled almost daily, and the US and 0:00:47.400 --> 0:00:50.840 China seemingly neck and neck in the race to dominate 0:00:51.040 --> 0:00:55.720 that transformative technology and reap the benefits. I mean, the 0:00:55.840 --> 0:00:57.760 challenge in that world is not that we have too 0:00:57.800 --> 0:01:00.560 little innovation, but too much, and how we got absorb 0:01:00.600 --> 0:01:03.600 it into the economy without blowing up everything that we 0:01:03.680 --> 0:01:08.080 hold deer. But another story, which we heard more often 0:01:08.120 --> 0:01:11.360 after the global financial crisis but has definitely not gone away, 0:01:11.600 --> 0:01:17.000 as one of stagnation of declining competition in large chunks 0:01:17.000 --> 0:01:20.480 of the economy, more and more meddling by governments. And 0:01:20.720 --> 0:01:22.800 of course we've heard that, particularly in Europe. But I 0:01:22.800 --> 0:01:25.880 think if you think about the rise of monopoly power 0:01:26.360 --> 0:01:29.240 more and more tariffs getting in the way of global trade. 0:01:29.240 --> 0:01:31.479 I mean, we've certainly seen all of that in Trump's 0:01:31.800 --> 0:01:35.720 American economy, and not to mention the more direct involvement 0:01:36.240 --> 0:01:38.920 in the affairs of business that the president and his 0:01:39.040 --> 0:01:43.479 cabinet have engaged in. So I was interested to try 0:01:43.520 --> 0:01:45.640 and tease out some of these themes and try and 0:01:45.640 --> 0:01:47.720 get a sense of are we in a very innovative 0:01:47.720 --> 0:01:51.040 era that means we're now about to have loads of 0:01:51.080 --> 0:01:56.920 growth or actually potentially running the risk of something more negative. 0:01:57.360 --> 0:02:00.440 I wanted to discuss it with Carl Benedict Free, who's 0:02:00.480 --> 0:02:02.920 the author of a book that came out last year Actually, 0:02:02.960 --> 0:02:08.120 How Progress Ends, Technology, Innovation, and the Face of Nations. 0:02:08.280 --> 0:02:11.800 He's also an associate professor at Oxford University and often 0:02:11.840 --> 0:02:14.760 writes for the Ft and other places. Car, thank you 0:02:14.840 --> 0:02:17.639 very much for joining Trump andomics. 0:02:17.639 --> 0:02:18.400 Good to be with you. 0:02:25.080 --> 0:02:27.480 Before we get into this, I think it's probably helpful 0:02:27.520 --> 0:02:31.240 for people to understand what the argument of your book is, 0:02:31.280 --> 0:02:38.440 and specifically your thesis around how technological and economic progress 0:02:38.919 --> 0:02:41.960 happen and the patterns that countries have tended to go through. 0:02:42.520 --> 0:02:44.040 So a key theme of the book is that the 0:02:44.040 --> 0:02:47.400 institutions need to just as technology moves along, right, and 0:02:47.440 --> 0:02:51.000 so you can grow for for a long period of 0:02:51.040 --> 0:02:55.080 time just by adopting a scaling technology invented elsewhere. And 0:02:55.160 --> 0:02:59.480 so the Soviet Union did that quite successfully over four decades. 0:02:59.560 --> 0:03:02.880 It took ample advantage of the Fortomotive Company's open door 0:03:03.040 --> 0:03:08.840 policy and transfer that knowledge to the Soviet Union build 0:03:09.080 --> 0:03:12.840 a vehicles industry of its own. And in the age 0:03:12.840 --> 0:03:16.400 of mass production, the Soviet system actually worked fairly well 0:03:16.480 --> 0:03:20.560 because you know, when technology is mature, when production is 0:03:20.600 --> 0:03:26.679 fairly standardized, then you can hold factory managers accountable essentially 0:03:26.760 --> 0:03:31.360 just by benchmarking performance. But as the returns to the 0:03:31.400 --> 0:03:35.680 mass production system petered out globally in the nineteen seventies 0:03:36.080 --> 0:03:39.520 and something new was needed for growth, that system did 0:03:39.600 --> 0:03:44.360 no longer work particularly well because when something is novel, 0:03:44.400 --> 0:03:46.800 when you're dealing with new technology, well, first of all, 0:03:46.800 --> 0:03:50.360 how do you benchmark performance? How do you hold factory 0:03:50.400 --> 0:03:54.800 managers accountable? And secondly, you know, that new thing that 0:03:54.920 --> 0:03:58.880 was needed for further growth was the computer revolution. To 0:03:58.960 --> 0:04:03.160 each Soviet contributeus were essentially non Why is that well, 0:04:03.560 --> 0:04:07.280 A big reason is that the Soviet Union provided very 0:04:07.320 --> 0:04:11.080 little room for decentralized exploration. And so if you were 0:04:11.080 --> 0:04:13.800 an engineer in the Soviet Union, you could go to 0:04:13.840 --> 0:04:16.599 the Red Arm and ask for funding. If they declined, 0:04:16.680 --> 0:04:20.560 well maybe two or three other options. If they declined, 0:04:20.640 --> 0:04:23.880 then your idea would die with you. And that's quite 0:04:23.920 --> 0:04:28.080 different from the American system of more decentralized finance, where 0:04:28.279 --> 0:04:31.200 you know, Besma Venture famously the client to invest in 0:04:31.240 --> 0:04:34.560 Google back in nineteen ninety nine. They probably regretted today, 0:04:34.960 --> 0:04:38.119 but it also illustrates that Google was not a safe 0:04:38.160 --> 0:04:41.360 bet at the time. Alta Vista and Yahoo they were 0:04:41.360 --> 0:04:47.200 dominating search, and so somebody you know, needed to invest 0:04:48.000 --> 0:04:52.520 to show that Google would actually catch on. And the 0:04:52.560 --> 0:04:54.919 fact that Besman didn't invest, though, didn't mean the end 0:04:54.960 --> 0:04:58.440 of Google because others stepped in. And so when you 0:04:58.520 --> 0:05:04.400 explore new technology, you need more people exploring different potential 0:05:04.520 --> 0:05:09.479 technological trajectories. So that's the exploration phase. But once you've 0:05:09.720 --> 0:05:12.760 settled on the product type, then you need to scale that. 0:05:12.920 --> 0:05:16.800 And that scaling in the Soviet Union did reasonably well. 0:05:17.200 --> 0:05:20.240 But when that runs into diminishing, it turns you need 0:05:20.279 --> 0:05:22.320 to get onto the new cycle. So you need to 0:05:22.360 --> 0:05:27.839 move from more centralized consolidation to more decentralized system again, 0:05:28.320 --> 0:05:33.080 and it's those shifts in institutions that make progress very 0:05:33.080 --> 0:05:33.920 hard to sustain. 0:05:35.000 --> 0:05:38.160 If you just look at the title of your book first, 0:05:38.480 --> 0:05:42.680 anyone looking at that, particularly this year, how progress ends 0:05:42.720 --> 0:05:45.719 technology innovation in the face of nations, you'd have to think, well, 0:05:46.760 --> 0:05:48.800 however it ends, it doesn't seem to be ending at 0:05:48.800 --> 0:05:53.080 the moment. So what's your take on the world we're 0:05:53.080 --> 0:05:55.200 looking at now? Are we at the end of progress 0:05:55.279 --> 0:05:57.120 or just the beginning of amazing progress? 0:05:57.800 --> 0:06:00.200 Well, the title of the book is not men to 0:06:00.240 --> 0:06:04.360 suggesting that progress is inevitably about to end. It's more 0:06:04.440 --> 0:06:07.640 a reflection of the fact that progress is unnatural. Right, 0:06:07.680 --> 0:06:10.440 If progress was inevitable, it would not have taken two 0:06:10.520 --> 0:06:14.200 hundred thousand years to having an industrial revolution. If progress 0:06:14.440 --> 0:06:16.840 was inevitable, the most of the world would be rich 0:06:16.880 --> 0:06:20.920 and prosperous. If progress was inevitable, Britain, the country where 0:06:20.920 --> 0:06:24.080 I live, would not have suffered two decades of productivity detegnation. 0:06:24.440 --> 0:06:27.760 So progress is clearly not inevitable, and it's clearly not 0:06:27.800 --> 0:06:33.800 inevitable even despite the acceleration we've seen in innovation. That dimension, right, 0:06:33.839 --> 0:06:35.720 So if you look at patenting, if you look at 0:06:35.720 --> 0:06:42.680 scientific publications, any measure of inventive output, all those indicators 0:06:42.720 --> 0:06:45.560 are up. Yet if you look at the economy as 0:06:45.600 --> 0:06:48.240 pointing in a very different direction, and even measures of 0:06:48.279 --> 0:06:52.719 research productivity and breaking through innovation are down. So we're 0:06:52.720 --> 0:06:57.960 getting more in terms of scientific and inventive output, but 0:06:58.040 --> 0:07:02.039 it seems that we're getting less transformational output. And I 0:07:02.080 --> 0:07:05.039 think that's important to keep in mind when we're looking 0:07:05.360 --> 0:07:10.520 at the potential impact of the economy on artificial intelligence 0:07:11.160 --> 0:07:14.880 as well, because in many ways the computer revolution was 0:07:14.960 --> 0:07:18.240 more transformative than the AI we have today. Right The 0:07:18.240 --> 0:07:22.200 computer and the Internet connected the best scientists and inventors 0:07:22.240 --> 0:07:26.440 around the world. It streamlined the research the process enormously. 0:07:26.800 --> 0:07:30.800 It gave us access to the weld store knowledge essentially 0:07:30.800 --> 0:07:34.320 in our pockets. And what do we get out of that? 0:07:34.800 --> 0:07:42.360 Basically a decade long productivity upsurge mostly confined to the 0:07:42.480 --> 0:07:46.360 United States. Now. I do think that AI will show 0:07:46.440 --> 0:07:50.480 up in the productivity statistics eventually, but the question is 0:07:51.000 --> 0:07:54.240 by how much and for how long? And so you know, 0:07:54.360 --> 0:07:59.720 any productivity upsurge, you'll take it. But I think if 0:07:59.760 --> 0:08:02.840 you believe that we are entering a new renaissance for 0:08:02.960 --> 0:08:09.280 economic growth. You're likely to be mistaken because AI is 0:08:09.640 --> 0:08:13.960 actually likely to give less of a productivity boost even 0:08:14.080 --> 0:08:17.280 the computer revolution. And the reason is this, with AI, 0:08:17.480 --> 0:08:22.120 still need verification. So AI automates, you know, a lot 0:08:22.160 --> 0:08:24.880 of knowledge work, but in the end of that, you 0:08:24.920 --> 0:08:27.200 need to verify the outputs, and so it's the time 0:08:27.320 --> 0:08:32.839 saving minus the time for verification. The computer revolution was different, right, 0:08:33.480 --> 0:08:37.800 you know, I could sit around and wait for hours, days, 0:08:37.840 --> 0:08:40.640 even weeks for new material to arrive for me to 0:08:40.679 --> 0:08:44.000 start my research, and the Internet gave me access to 0:08:44.080 --> 0:08:47.680 that instantaneously, and so it's automated downtime. AI is not 0:08:47.840 --> 0:08:51.040 automating that downtime. It's automating the production, and you still 0:08:51.040 --> 0:08:52.720 need a verification as well. 0:08:53.080 --> 0:08:56.200 I'm sure we'll get onto some of the implications of 0:08:56.400 --> 0:08:59.760 AI and also sort of the bearing of some of 0:08:59.800 --> 0:09:01.880 your research on that, because you sort of look back 0:09:01.920 --> 0:09:04.640 at thousands of years of what has produced innovation and 0:09:04.679 --> 0:09:08.120 then what's then translated that into growth. But just what 0:09:08.160 --> 0:09:12.720 you said about that sort of disconnect around, we've seen 0:09:12.760 --> 0:09:15.720 more invention at some level, lots more invention, lots more 0:09:15.760 --> 0:09:19.320 patenting and everything else, and yet seemingly at the micro level, 0:09:19.360 --> 0:09:22.480 at the real level of the economy, still quite a 0:09:22.480 --> 0:09:27.280 lot of low productivity and lack of translation of that 0:09:27.559 --> 0:09:32.400 technology into productivity. And thinking about that contrast between the 0:09:32.520 --> 0:09:35.000 US and China, there was a story that was told 0:09:35.040 --> 0:09:38.000 for many years that probably is still being told about 0:09:38.440 --> 0:09:44.520 the innate advantages of the US in technological progress that 0:09:44.679 --> 0:09:47.199 was always around. You know, China might be very good 0:09:47.200 --> 0:09:51.239 at copying things, but when it comes to actual innovation 0:09:51.640 --> 0:09:55.480 at the frontier, the US is always going to have 0:09:55.640 --> 0:09:59.520 an advantage over a sort of centralized system China. There's 0:09:59.559 --> 0:10:03.120 quite a lot support for that in your book, But 0:10:03.200 --> 0:10:05.160 I guess if you look all around, you would have 0:10:05.240 --> 0:10:07.199 to sort of say, well, maybe Ai is going to 0:10:07.240 --> 0:10:09.400 be the exception, because they seem to be getting pretty 0:10:09.440 --> 0:10:13.439 close to the frontier on AI without having the kind 0:10:13.480 --> 0:10:15.959 of decentralized system that the US has. 0:10:16.440 --> 0:10:18.760 Well, I think, first of all, though clearly is innovation 0:10:18.840 --> 0:10:21.600 in China, But China is also a country of one 0:10:21.640 --> 0:10:25.040 point four billion people, so it would be absolutely extraordinary 0:10:25.280 --> 0:10:30.120 if there was no innovation going on in China whatsoever. 0:10:30.559 --> 0:10:33.920 But in addition to that, though China is not as 0:10:33.960 --> 0:10:38.800 centralized as is commonly believed, and so I like the 0:10:38.840 --> 0:10:42.280 Soviet Union, which also had one party system, where every 0:10:42.320 --> 0:10:46.400 industry was managed centrally from Moscow, whether it was railroads 0:10:46.480 --> 0:10:52.000 or steel. The Chinese economy is much more decentralized, and 0:10:52.440 --> 0:10:58.480 provinces and provincial governors and mayors have much greater autonomy 0:10:58.840 --> 0:11:03.560 than the counterparts would have had. And so what you 0:11:03.640 --> 0:11:11.720 have in China system where provincial governors are competing against 0:11:11.720 --> 0:11:17.360 each other based on often growth targets for promotion inside 0:11:17.800 --> 0:11:21.400 the one party system, and so what that creates is 0:11:21.520 --> 0:11:28.800 essentially a tournament of political competition that is very hard 0:11:28.960 --> 0:11:33.720 to replicate anywhere else. And so Europe does industrial policy, 0:11:33.960 --> 0:11:36.240 it often tends to be anti competitive, right where you 0:11:36.280 --> 0:11:42.319 German rearmament essentially means plowing funds into Rhymtal. In China, 0:11:42.760 --> 0:11:45.000 on the other hand, industrial policy can often be pro 0:11:45.040 --> 0:11:49.920 competitive because you have provinces seeding new firms that are 0:11:49.920 --> 0:11:54.600 competing against each other, and so I think that's often 0:11:54.840 --> 0:11:59.679 underappreciated part of the Chinese economy. That said, though, when 0:11:59.720 --> 0:12:03.760 you look get you know the firms which are leading 0:12:03.800 --> 0:12:09.800 in innovation in China, it's mostly startups, it's mostly privately 0:12:10.040 --> 0:12:15.200 funded firms, it's often foreign funded firms as well, and 0:12:15.240 --> 0:12:19.240 so in that sense, China is not that different from 0:12:19.720 --> 0:12:23.000 either Europe and the United States. What is different is 0:12:23.000 --> 0:12:25.720 that it doesn't have the rule of law, and so 0:12:26.800 --> 0:12:31.280 building political connections is more important in China because you 0:12:31.360 --> 0:12:34.200 need to have a seat of the table when priorities changed, 0:12:34.240 --> 0:12:38.479 and priorities changed from time to time. But perhaps somewhat paradoxically, 0:12:38.520 --> 0:12:41.080 what people believed in the two thousands that China would 0:12:41.080 --> 0:12:44.040 could become more like the United States, the opposite seems 0:12:44.040 --> 0:12:48.360 to be happening, and the United States today looks more 0:12:48.400 --> 0:12:52.199 like the political capitalism and that you have in China. 0:12:52.559 --> 0:12:54.719 And you know, being on the right side of the 0:12:54.760 --> 0:12:59.719 current administration is important, which is probably one of the 0:12:59.720 --> 0:13:02.160 reasons since that firms like Open the Eye are saying 0:13:02.200 --> 0:13:06.280 that we are open to the government taking five percent 0:13:06.320 --> 0:13:08.839 stake in the company, because if they are on your side, 0:13:08.920 --> 0:13:13.120 well then you're on the right side of things from 0:13:13.160 --> 0:13:14.320 your perspective at least. 0:13:14.520 --> 0:13:18.280 That's a quite important theme in your book, which is 0:13:18.440 --> 0:13:25.240 that ultimately has undermined in the past a country's position 0:13:25.400 --> 0:13:28.760 at the sort of cutting edge of innovation but also 0:13:28.880 --> 0:13:32.000 of growth. And you have the example of the UK 0:13:32.160 --> 0:13:36.319 after the industry revolution is that you have vested interests 0:13:36.400 --> 0:13:41.719 build up which slowly kind of resists disruption and competition 0:13:42.120 --> 0:13:45.600 and new entrance to the tow industries, which you then 0:13:45.800 --> 0:13:49.079 you know, gradually lose your innovative edge. You seem to 0:13:49.120 --> 0:13:53.080 see scope for that happening both in the US and 0:13:53.160 --> 0:13:54.920 in China at the moment, But I guess for very 0:13:54.960 --> 0:13:55.720 different reasons. 0:13:56.480 --> 0:13:58.160 Yeah, And I think, you know, I think that's the 0:13:58.240 --> 0:14:00.599 natural order of things, right. So if you look at historically, 0:14:00.640 --> 0:14:04.679 the leaders in bicycles didn't become leaders in automotive, although 0:14:04.720 --> 0:14:08.439 they try, right. Legacy media companies did not lead the 0:14:08.480 --> 0:14:12.120 social media revolution, The legacy car companies did not lead 0:14:12.120 --> 0:14:15.800 in electric vehicles. The legacy retailers did not lead in 0:14:15.880 --> 0:14:19.440 e commerce. So it's a clear pattern, the pattern that 0:14:19.520 --> 0:14:24.280 it's new firms that tend to develop new technology and 0:14:24.600 --> 0:14:29.520 new kinds of industries, and old industries have an incentive 0:14:29.600 --> 0:14:33.200 or ever incentive to prevent that sort of competition. And 0:14:33.280 --> 0:14:35.480 so that's why we're see in the rise of killer 0:14:35.520 --> 0:14:39.720 acquisitions in the United States, for example, whereby incumbents buy 0:14:39.800 --> 0:14:43.600 up promising startups just to shut them down. That's why 0:14:43.600 --> 0:14:46.280 we see a revolving door between the US Patent and 0:14:46.320 --> 0:14:51.840 Trademark Office and some incumbents, whereby you know, patent examiners 0:14:51.960 --> 0:14:55.000 grant them low quality patents and then take on jobs 0:14:55.600 --> 0:15:00.560 for these firms in return. And that's all that. It's 0:15:00.680 --> 0:15:04.640 creating barriers to entry for new companies, and it helps 0:15:04.680 --> 0:15:08.400 explain why we see in a decline in business dynamis 0:15:08.760 --> 0:15:12.000 despite the fact that every technology from the personal computer, 0:15:12.160 --> 0:15:15.000 the internet, the cloud, and now ai will have made 0:15:15.040 --> 0:15:17.440 it much cheaper to set up a company and operate 0:15:18.200 --> 0:15:20.600 a firm, and yet we're see less entry and so 0:15:20.680 --> 0:15:24.720 those barriers are clearly at work in the US and 0:15:24.760 --> 0:15:28.520 they were at work in China as well, although in 0:15:28.600 --> 0:15:35.800 China you have the added component that the priorities of 0:15:35.840 --> 0:15:39.960 the CCP has changed over the past fifteen years or 0:15:39.960 --> 0:15:47.600 so from basically economic targets to targets that concern political 0:15:47.680 --> 0:15:54.240 means around self sufficiency and national security, common prosperity and 0:15:54.280 --> 0:15:58.000 so on. And what that means in the Chinese case 0:15:58.120 --> 0:16:03.760 is that that creates it greater reliance on state owned enterprises, 0:16:04.280 --> 0:16:07.200 because private firms are for the most part less keen 0:16:07.240 --> 0:16:13.400 on pursuing national non economic objectives, and by any measure, 0:16:13.880 --> 0:16:19.040 state owned enterprises in China have been less innovative and 0:16:19.240 --> 0:16:31.080 less productive, and that I think is unlikely to change. 0:16:26.040 --> 0:16:38.360 And one set of sort of arguments against or points 0:16:38.360 --> 0:16:42.560 against your thesis would be sort of in the realm 0:16:42.640 --> 0:16:46.240 of this time is different, or that AI might be 0:16:46.320 --> 0:16:48.840 a different kind of technology. You make the example of 0:16:48.880 --> 0:16:54.640 the bicycles to cars and another sort of classic technological shifts. 0:16:55.280 --> 0:16:58.640 And you know, one argument that you could make just 0:16:58.760 --> 0:17:02.000 looking at the way has evolved over the last few 0:17:02.040 --> 0:17:04.440 years and even the last few months, is it seems 0:17:04.480 --> 0:17:08.679 like scale matters more than anything else in this the 0:17:08.720 --> 0:17:12.639 development of this technology. And whereas in the past we 0:17:12.720 --> 0:17:14.520 might have thought that it was going to be about 0:17:14.600 --> 0:17:20.240 different innovations around different ways of doing things, the big 0:17:20.280 --> 0:17:23.159 developments in AI seem to be coming just from throwing 0:17:23.240 --> 0:17:28.120 a lot of resource at llm's and you know it's 0:17:28.160 --> 0:17:30.960 going to and that the gains go to the country 0:17:31.000 --> 0:17:34.560 that has the most energy and the most chips, and 0:17:34.600 --> 0:17:37.600 at the moment, certainly the most energy is China doesn't 0:17:37.640 --> 0:17:41.120 currently have the most chips. But doesn't seem to suggest 0:17:41.119 --> 0:17:42.960 that you need a lot of innovation or sort of 0:17:43.320 --> 0:17:46.800 new entrance, because actually the new entrants can't necessarily afford 0:17:47.040 --> 0:17:50.440 to put in all these enormous fixed costs. So what's 0:17:50.480 --> 0:17:53.040 the chance that this is just a different kind of technology. 0:17:53.119 --> 0:17:56.240 It's going to be innovating itself if it's given enough 0:17:56.359 --> 0:17:57.320 resource to do it. 0:17:57.760 --> 0:18:00.280 So, if what you've said is correct and clear, what 0:18:00.320 --> 0:18:03.520 I've written and said is wrong, right, And so I 0:18:03.560 --> 0:18:06.639 don't believe that that is the case though. And so 0:18:07.160 --> 0:18:10.040 you know, if the world was just a static distribution 0:18:10.080 --> 0:18:12.960 of events, you could probably you know, brute force things, right, 0:18:13.000 --> 0:18:16.480 So more you know, more compute, more data world, you know, 0:18:16.520 --> 0:18:18.600 eventually get you there, right. But the world is not 0:18:18.640 --> 0:18:21.800 just a static distribution of events. It's changing all the time, right, 0:18:22.200 --> 0:18:26.600 And so you know, my job today may be very 0:18:26.640 --> 0:18:30.760 different from my job tomorrow. Not all work is like that. 0:18:30.840 --> 0:18:32.960 And so there's a lot of things that static and 0:18:33.080 --> 0:18:35.600 you can automate that, but you know a lot of 0:18:35.800 --> 0:18:39.159 things also requires some degree of resilience, right, And so 0:18:39.200 --> 0:18:42.040 if you use AI to manage supply chains, that might, 0:18:42.119 --> 0:18:43.639 you know, work quite well, and then all of a 0:18:43.680 --> 0:18:46.159 sudden you have pandemic and you don't know what on 0:18:46.200 --> 0:18:49.040 earth is going on. And we see this, you know, 0:18:49.119 --> 0:18:52.359 even in more confined spaces, right, So most people know 0:18:52.720 --> 0:18:58.359 of AlphaGo that beat licid all forty one back in 0:18:59.160 --> 0:19:02.760 twenty six, and so you know, by that time already 0:19:03.520 --> 0:19:08.200 AI had achieved super human performance in GO. Few people 0:19:08.320 --> 0:19:11.600 know that actually, just a couple of years ago, human 0:19:11.640 --> 0:19:17.480 amateurs using standard computers be the best available GO programs 0:19:18.040 --> 0:19:22.160 quite easily by exposing them to new positions, new concepts 0:19:22.160 --> 0:19:25.399 that they would not have encountered in training. And so 0:19:25.480 --> 0:19:28.080 that rises a sort of fundamental question of you know, 0:19:28.520 --> 0:19:31.560 even in cases where we achieve super human performance, we 0:19:31.640 --> 0:19:34.120 cannot be sure if that's actually going to be true 0:19:34.160 --> 0:19:39.480 tomorrow when circumstances change. And so, you know, humans we 0:19:39.520 --> 0:19:43.879 are capable of learning from just a few examples. We 0:19:43.920 --> 0:19:47.840 are very very data efficient, and I think you will 0:19:47.880 --> 0:19:51.480 have to get to AI that is capable of that, 0:19:52.119 --> 0:19:56.640 and that will need some innovation. And so right now 0:19:56.760 --> 0:20:00.480 we don't really know what the path forward in AI 0:20:00.600 --> 0:20:03.600 looks like. It may be that's you know, large language models, 0:20:03.640 --> 0:20:05.960 it's the future of AI. It may be that small 0:20:06.040 --> 0:20:08.880 language models, it might be world models, it might be 0:20:08.920 --> 0:20:12.320 something entirely different. And so I think what we need 0:20:12.400 --> 0:20:15.679 is greater data efficiency, and that's clearly not something we're 0:20:15.720 --> 0:20:18.920 going to get through just through scaling. AI is still 0:20:19.000 --> 0:20:23.439 you know, waiting for what I call the separate condenser moment, 0:20:23.600 --> 0:20:28.479 because you know, this first Industrial Revolution, early on steam engines, 0:20:28.520 --> 0:20:31.439 they were tremendously energy and efficient. They were basically just 0:20:31.560 --> 0:20:34.639 used to drain coal mines. They couldn't really be used 0:20:34.680 --> 0:20:37.359 for anything else, and it took the separate condenser to 0:20:37.400 --> 0:20:40.679 make them energy efficient for them to be applied to 0:20:40.840 --> 0:20:45.879 transportation later on railroads and steamships, et cetera. And I 0:20:45.880 --> 0:20:48.280 think AI is still waiting for that moment, and that's 0:20:48.520 --> 0:20:51.280 a question of you know, further innovation, not just scaling. 0:20:51.520 --> 0:20:55.399 It's also it's a very good documentary about the AlphaGo experience, 0:20:55.720 --> 0:20:59.160 and it was this sort of classic example that stayed 0:20:59.200 --> 0:21:02.480 in people's mind of like when the humans lost gets 0:21:02.960 --> 0:21:05.760 this new technology. I was surprised to read in the 0:21:05.760 --> 0:21:08.400 book the idea that you actually had had a fight 0:21:08.480 --> 0:21:13.399 back by people who were using techniques that AlphaGo had 0:21:13.440 --> 0:21:16.679 not been able to surmise from it from what it 0:21:16.760 --> 0:21:19.359 was learning from, And sort of interesting to me that 0:21:19.400 --> 0:21:21.600 people are determined to think that humans are going to 0:21:21.640 --> 0:21:25.960 lose this race rather than jumping on those kind of examples. 0:21:26.640 --> 0:21:30.000 So it seems like you would you know, Again, I 0:21:30.240 --> 0:21:33.760 mentioned at the start there's this kind of dizzying array 0:21:33.800 --> 0:21:37.560 of new models of AI, and then debates about where, 0:21:37.760 --> 0:21:41.720 you know, how close the US is to China when 0:21:41.720 --> 0:21:44.639 you're trying to think about AI's impact on the world 0:21:44.720 --> 0:21:48.000 and whether the US is ahead of China or behind China. 0:21:48.200 --> 0:21:50.160 Is that even the right way to think about it 0:21:50.560 --> 0:21:53.919 when we look at the sheer power of these models 0:21:53.760 --> 0:21:56.080 as a gauge to that, or should we be looking 0:21:56.119 --> 0:21:57.840 at other aspects of their economies. 0:21:58.680 --> 0:22:03.600 So unless we get to a stage where some firm, 0:22:04.119 --> 0:22:08.240 you know, some company gets onto curve where it really, 0:22:08.280 --> 0:22:12.200 you know, just pulls away from the rest, I don't 0:22:12.240 --> 0:22:17.040 think you know whether the US or China or you know, 0:22:17.280 --> 0:22:21.800 open R or and Tropic or Google is three months 0:22:21.800 --> 0:22:26.879 ahead or not. If one place really pulls ahead, that 0:22:26.960 --> 0:22:30.680 could have a meaningful impact, But it's not clear that 0:22:30.680 --> 0:22:33.040 that's going to be within the sort of space of 0:22:33.119 --> 0:22:36.639 large language models. It might be something else. And then 0:22:36.680 --> 0:22:39.639 there's a question of adoption. And obviously, in the end 0:22:39.680 --> 0:22:41.760 of the day, the use of a technology is what 0:22:41.960 --> 0:22:45.879 drives productivity, but they're also the question is adoption for 0:22:45.960 --> 0:22:49.119 what if people adopted for email, it's not going to 0:22:49.200 --> 0:22:52.560 drive growth in a meaningful way, but you know, it may 0:22:53.040 --> 0:22:55.840 look good in sense that the adoption rate is high. 0:22:55.960 --> 0:22:59.800 But if people adopt it for scientific research in ways 0:22:59.840 --> 0:23:04.199 that develop new products and new technologies, that's obviously a 0:23:04.240 --> 0:23:07.400 different matter. And I think there, you know, incentives matter 0:23:07.480 --> 0:23:10.000 a lot, and I think that has some bearing on 0:23:10.080 --> 0:23:14.960 why we're in this sort of global productivity stagnation since 0:23:14.960 --> 0:23:17.720 the computer revolution. Really, so what you really see with 0:23:18.160 --> 0:23:22.960 the computer era is that inventors and scientists have taken 0:23:23.000 --> 0:23:25.480 on more projects since then. So when you get a 0:23:25.520 --> 0:23:27.639 new powerful productivity tool, right, you can be one of 0:23:27.960 --> 0:23:30.840 two things. You can either use it to drill deeper 0:23:30.960 --> 0:23:33.280 or dig deeper, or you could use it to your 0:23:33.400 --> 0:23:37.760 drill more holes. And if you do more projects, just 0:23:37.880 --> 0:23:40.280 drill more holes. In the end of the day, your 0:23:40.320 --> 0:23:45.359 attention is going to be more thinly spread across multiple projects, 0:23:45.640 --> 0:23:48.439 and as a result of that, you are actually less 0:23:48.600 --> 0:23:52.159 likely to make a breakthrough at any given time. We 0:23:52.200 --> 0:23:54.680 see that in the data. AI seems to have had 0:23:54.720 --> 0:23:58.240 the same effect. AI means that we can do more things, 0:23:58.560 --> 0:24:00.760 but people seem to be using it to do more 0:24:00.840 --> 0:24:05.000 things rather than digging deeper, and so in academia the 0:24:05.040 --> 0:24:07.600 incentive is publish a parish, and so we shouldn't be 0:24:07.760 --> 0:24:11.480 surprised if people use it to produce more output rather 0:24:11.600 --> 0:24:15.600 than you know, spending the next ten years maybe producing nothing, 0:24:15.920 --> 0:24:18.240 by coming out with a real breakthrough by the end 0:24:18.280 --> 0:24:18.879 of that period. 0:24:19.760 --> 0:24:25.280 Finishing this line of thought, if you're country, as most 0:24:25.320 --> 0:24:30.720 countries are, that are not home to companies that are 0:24:30.760 --> 0:24:35.120 absolutely at the forefront of producing this technology, probably your 0:24:35.119 --> 0:24:38.280 government is saying adoption is key. Certainly we hear this 0:24:38.359 --> 0:24:42.240 in UK, we hear this in Europe, and to your point, 0:24:42.280 --> 0:24:44.040 you know, the most important thing is going to be 0:24:44.080 --> 0:24:46.800 the way we adopt this technology, not whether we happen 0:24:46.880 --> 0:24:51.080 to own this or that piece of it. You know, 0:24:51.119 --> 0:24:56.680 what are the lessons from your analysis for governments that 0:24:56.840 --> 0:25:01.840 want to try and make the most of these technological 0:25:01.880 --> 0:25:05.600 advances and translate them into growth. 0:25:06.400 --> 0:25:09.600 So I think if you're behind the frontier, which is 0:25:09.880 --> 0:25:14.159 inevitably true for most places, and you know that was 0:25:14.200 --> 0:25:17.200 true d in the Second Industrial Revolution the computer revolution 0:25:17.800 --> 0:25:20.240 as well, the best thing you can do is trying 0:25:20.280 --> 0:25:24.119 to adopt technology invented elsewhere. And then the question is, 0:25:24.160 --> 0:25:26.600 obviously how easy is that and so during the post 0:25:26.640 --> 0:25:31.359 war period, the United States had an explicit policy through 0:25:31.520 --> 0:25:36.199 martial aid, where it shared technology with its allies, and 0:25:36.240 --> 0:25:39.560 that's contributed, i think to a meaningful degree to the 0:25:39.600 --> 0:25:46.480