WEBVTT - AI Is Being Built to Replace You—Not Help You 0:00:02.720 --> 0:00:09.560 Bloomberg Audio Studios, Podcasts, radio News. On our current path, 0:00:09.960 --> 0:00:12.400 this handful of people will decide what the future of 0:00:12.440 --> 0:00:17.159 AI is, and the best way to counter balance that 0:00:17.680 --> 0:00:20.960 is to have a vision that's different and hopefully better 0:00:21.040 --> 0:00:26.920 for society. 0:00:31.080 --> 0:00:33.720 Welcome to Trumpnomics, the podcast that looks at the economic 0:00:33.760 --> 0:00:36.479 world of Donald Trump, how he's already shaped the global 0:00:36.520 --> 0:00:38.400 economy and modern earth is going to. 0:00:38.400 --> 0:00:39.320 Happen next. 0:00:41.200 --> 0:00:43.880 Well, we talk about the big forces affecting our economy 0:00:43.920 --> 0:00:45.720 in the broader world on this show. And there's no 0:00:45.760 --> 0:00:49.519 bigger topic in economic policy and general discussion these days 0:00:49.560 --> 0:00:53.200 than the impact of AI. We've talked at different times 0:00:53.200 --> 0:00:57.200 about the consequences for jobs, inflation, interest rates, and where 0:00:57.200 --> 0:01:00.600 the policy makers, let alone ordinary people, ready for any 0:01:00.640 --> 0:01:04.160 of it. Well, my guest this week, Darren Asimoglu, famously 0:01:04.200 --> 0:01:06.880 takes the long view on these matters. A recipient of 0:01:06.880 --> 0:01:09.920 the Nobel Prize for Economics in twenty twenty four, he 0:01:10.000 --> 0:01:13.000 probably wrote the single most widely read book of economic 0:01:13.120 --> 0:01:16.880 history of recent times, Why Nations Fail, And more recently 0:01:16.959 --> 0:01:20.800 he wrote with Simon Johnson, Power and Progress, Our thousand 0:01:20.840 --> 0:01:24.880 years struggle over technology and prosperity, and I talked to 0:01:24.959 --> 0:01:27.679 him about that book and the lessons for this AI 0:01:27.760 --> 0:01:31.120 revolution a while back in the summer of twenty twenty three, 0:01:31.360 --> 0:01:33.720 but given everything that's been going on, I wanted to 0:01:33.760 --> 0:01:37.120 have him back to see whether he saw anything in 0:01:37.280 --> 0:01:41.360 the new waves of AI that we've had since twenty 0:01:41.400 --> 0:01:44.119 twenty three, particularly in the last few months, whether he'd 0:01:44.120 --> 0:01:47.800 seen anything to make him change his view on either 0:01:47.840 --> 0:01:50.560 how fast this technology is going to change our economy 0:01:50.960 --> 0:01:53.600 or how well placed we are to get the best 0:01:53.600 --> 0:01:56.000 out of it. Darren, thanks very much for coming back 0:01:56.040 --> 0:01:57.120 on to this podcast. 0:01:57.720 --> 0:02:02.560 Thank you. Stephanie's my pleasure to be here. 0:02:04.520 --> 0:02:08.240 We will get into some of the sort of key 0:02:08.320 --> 0:02:10.840 dimensions of this in a minute, but I should just 0:02:10.880 --> 0:02:12.960 get a sense from you. I mean, we're talking now 0:02:13.080 --> 0:02:17.400 mid March twenty twenty six. There's been so much chatter, 0:02:17.680 --> 0:02:19.840 and I suspect many people listening have had their own 0:02:19.880 --> 0:02:24.720 real life experience now of the development of all the 0:02:24.760 --> 0:02:27.400 different forms of AI, and particularly the sort of agentic 0:02:27.480 --> 0:02:30.600 AI that we talk about. Are you, in a kind 0:02:30.600 --> 0:02:35.079 of very broad sense reassessing how fast or how fundamentally 0:02:35.120 --> 0:02:36.440 this is going to change our world? 0:02:37.280 --> 0:02:42.720 Yeah? Every day, I think the underlying technology is changing 0:02:42.880 --> 0:02:46.519 faster than what I would have predicted, what many would 0:02:46.560 --> 0:02:51.600 have predicted eight and a half ago. So, especially with 0:02:52.680 --> 0:02:57.000 the recent developments in agentic AI, especially led by Anthropic, 0:02:58.480 --> 0:03:01.919 there is a real possible ability that these tools can 0:03:02.000 --> 0:03:07.200 be broadly useful in what people do. There is still 0:03:07.280 --> 0:03:10.880 a lot of uncertainty, however. First we are not seeing 0:03:11.000 --> 0:03:17.000 any of the prepackaged, easy to use, reliable applications think 0:03:17.040 --> 0:03:20.920 of it as the Microsoft Words or Microsoft Offices of 0:03:21.320 --> 0:03:25.160 AI that can be used across a broad range of 0:03:25.240 --> 0:03:28.680 occupations or in some specific occupations. Those are not around yet. 0:03:29.120 --> 0:03:34.680 There is still uncertainty about whether there will be bottlenecks 0:03:34.720 --> 0:03:39.760 in reaching higher reliability and higher judgment. There is every 0:03:40.440 --> 0:03:45.200 evidence that there is a lot of rapid progress, but 0:03:45.520 --> 0:03:50.200 there are some weaknesses in these models that are still persistent, 0:03:50.960 --> 0:03:54.280 and I don't just mean hallucinations, but lack of a 0:03:54.360 --> 0:03:57.440 deep understanding. They don't seem to have a conceptual framework, 0:03:57.480 --> 0:04:01.240 they don't understand the context, and they cannot reason at 0:04:01.400 --> 0:04:06.080 multiple levels of abstraction about a problem. Yet so those 0:04:06.240 --> 0:04:08.480 may be overcome, and I think many of those are 0:04:08.520 --> 0:04:12.120 going to be important in dealing with edge cases in 0:04:12.160 --> 0:04:19.440 many occupations. So wholesale automation of occupations is still not 0:04:19.680 --> 0:04:22.960 something we're going to see right away, but some people 0:04:23.160 --> 0:04:24.960 swear that we're going to see it in one year 0:04:25.040 --> 0:04:27.960 or two years, three years. So there's a lot of uncertainty. 0:04:28.040 --> 0:04:33.279 But let me make one thing clear. If we do 0:04:33.440 --> 0:04:39.000 not up our game about both how we regulate these 0:04:39.040 --> 0:04:43.039 models and how we actually develop them, there could be 0:04:43.040 --> 0:04:45.000 a huge amount of damage to society. 0:04:45.640 --> 0:04:48.719 I think that's very helpful because there's two elements of 0:04:48.760 --> 0:04:51.520 this where there's obviously, as you say, there's a lot 0:04:51.520 --> 0:04:54.479 of uncertainty, and there's a wide range of opinion. If possible, 0:04:54.480 --> 0:04:56.240 I'm going to try and separate them, but obviously they 0:04:56.320 --> 0:04:58.839 merge into each other. One is this question of the 0:04:58.880 --> 0:05:01.240 pace of change, faster company is really going to be 0:05:01.279 --> 0:05:04.479 able to change their practices or capture those productivity improvements. 0:05:04.960 --> 0:05:07.680 And then the second is which you've just highlighted, is 0:05:07.720 --> 0:05:09.960 how well are we positioned to make the best of this, 0:05:10.160 --> 0:05:13.159 not just to get all the productivity growth, but to 0:05:13.200 --> 0:05:15.400 make sure it's actually positive for most of the population, 0:05:16.000 --> 0:05:18.080 not just a few. And you highlighted in the twenty 0:05:18.160 --> 0:05:20.080 twenty three book, you know that none of that was 0:05:20.120 --> 0:05:22.839 automatic in the case of the industrial revolution, and we 0:05:22.880 --> 0:05:25.200 may have to do it much faster this time. Just 0:05:25.279 --> 0:05:29.280 focusing on this speed question, there's a Citrini report which 0:05:29.279 --> 0:05:31.640 there's been a little mini industry in sort of debunking 0:05:31.680 --> 0:05:35.320 this research report which sort of went viral because it 0:05:35.360 --> 0:05:37.760 captured some elements of this sort of faster pace that 0:05:37.800 --> 0:05:39.640 we were seeing. I think you wrote a paper, you know, 0:05:39.680 --> 0:05:43.120 the basic Macroeconomics of AI. In the debate, I would 0:05:43.120 --> 0:05:47.080 say that you're fairly low key about the pace of change, 0:05:47.120 --> 0:05:49.320 the extent of change in any given year. I think 0:05:49.320 --> 0:05:53.120 you said at most a few percent over ten years, 0:05:53.160 --> 0:05:55.920 so maybe even a fraction of a percent of productivity 0:05:55.960 --> 0:05:59.400 growth overall productivity growth a year. Would you stand by 0:06:00.080 --> 0:06:03.600 that basic assessment today or do you think maybe the gains, 0:06:03.880 --> 0:06:06.239 just the purebreductivity gains could be a bit faster. 0:06:06.920 --> 0:06:09.560 I think they could be a bit faster. There has 0:06:09.640 --> 0:06:17.719 been faster change of the capabilities of the foundation models. However, 0:06:18.720 --> 0:06:22.680 it would still require some big breakthroughs, especially at the 0:06:22.720 --> 0:06:29.840 application layer. So the bottom line of that paper was 0:06:29.920 --> 0:06:34.200 to point out how we can get a fairly simple 0:06:34.760 --> 0:06:39.920 understanding of the constituent parts of the contribution of AI 0:06:40.640 --> 0:06:46.360 to productivity and GDP growth, And that comes from realizing 0:06:46.560 --> 0:06:50.160 that the GDP contribution of AI is nothing other than 0:06:50.600 --> 0:06:52.920 what fraction of tasks are going to be taken over 0:06:53.200 --> 0:06:57.160 or completely transformed by AI in the economy times the 0:06:57.200 --> 0:07:00.960 average productivity gain or average cost savings in these tasks. 0:07:01.640 --> 0:07:04.640 So that's the calculation that I did with the available 0:07:04.720 --> 0:07:08.960 evidence in twenty twenty two. But even then a lot 0:07:08.960 --> 0:07:13.760 of people took issue at how I interpreted the data, etc. 0:07:14.120 --> 0:07:17.400 So you could boost some of the numbers that I have, 0:07:17.640 --> 0:07:20.640 which were that about five percent of the whole economy 0:07:20.680 --> 0:07:24.520 will be taken by AI within ten years, by twenty 0:07:24.560 --> 0:07:29.400 thirty or thereabouts, and that that would lead to about 0:07:29.640 --> 0:07:34.960 twenty five percent cost savings or relative to labor costs 0:07:35.040 --> 0:07:39.440 that firms used to spend on the same tasks. Now 0:07:39.480 --> 0:07:44.960 you can boost my numbers by increasing either or both 0:07:45.000 --> 0:07:48.000 of these quantities. So you can say, no, no, it's 0:07:48.040 --> 0:07:51.320 not five percent, it's going to be twenty percent of 0:07:51.360 --> 0:07:53.680 the economy that AI is going to take over, in 0:07:53.720 --> 0:07:56.880 which case you would quadruple my numbers. Or you could 0:07:56.880 --> 0:08:00.280 say it's not twenty percent cost savings, but it's going 0:08:00.320 --> 0:08:03.040 to lead to thirty percent or forty percent cost savings. 0:08:03.440 --> 0:08:06.200 After all, you know, comp lead to three hundred percent 0:08:06.240 --> 0:08:09.480 cost savings because you know labor wasn't very expensive anyway. 0:08:09.640 --> 0:08:12.920 But you see the elbow room to do that kind 0:08:12.920 --> 0:08:14.720 of thing. But you're not going to come up with 0:08:14.800 --> 0:08:18.200 revolutionary numbers here. And part of the problem here is 0:08:18.200 --> 0:08:22.600 that we are right now, and that was the case 0:08:22.600 --> 0:08:25.080 two years ago and continues to be the case. We 0:08:25.160 --> 0:08:29.160 are right now focusing on AI as an automation tool, 0:08:29.240 --> 0:08:33.120 as a tool to replace workers. That's not the best 0:08:33.200 --> 0:08:36.360 way of using AI. The best way of using AI 0:08:36.640 --> 0:08:39.719 is to try to complement workers so that they can 0:08:39.760 --> 0:08:42.720 do new things, they can perform new tasks, they can 0:08:43.280 --> 0:08:48.920 increase their sophistication level, and also respond to challenges in 0:08:48.960 --> 0:08:52.520 the world economy from globalization, from aging, from climate change 0:08:52.880 --> 0:08:56.280 by creating new goods and services, new organizations and so on. 0:08:56.480 --> 0:08:58.679 If we do that, I think I would be more 0:08:58.720 --> 0:09:02.000 optimistic about the future away. And it's one of the 0:09:02.040 --> 0:09:04.720 aspects of the wisdom gap that we have right now. 0:09:05.000 --> 0:09:07.280 We don't know how to regulate existing models, and we 0:09:07.320 --> 0:09:10.160 are not really focusing on what we can do best 0:09:10.440 --> 0:09:13.800 with these models. And that's both for productivity and social consequences. 0:09:13.880 --> 0:09:16.320 So I've just made the productivity case that we could 0:09:16.320 --> 0:09:19.880 actually get better productivity consequences. But actually the social consequences 0:09:19.920 --> 0:09:23.000 are even starker. If we displace people, if we say 0:09:23.040 --> 0:09:25.640 displaced twenty percent of the population from their jobs and 0:09:25.640 --> 0:09:28.760 they remain unemployed or they go to lower quality, lower 0:09:28.800 --> 0:09:32.000 pay jobs, our democracy is not going to survive. We're 0:09:32.040 --> 0:09:37.760 already struggling to make our democratic system. Yeah, we're not 0:09:37.760 --> 0:09:41.640 doing that well. And if we put another huge shock 0:09:41.679 --> 0:09:43.760 on top of that, I'm not very optimistic. 0:09:43.880 --> 0:09:47.120 So beware and just thinking about the economics of what 0:09:47.160 --> 0:09:51.000 you're saying, and also thinking about what captured people's imaginations 0:09:51.040 --> 0:09:53.760 about that Suitrinian report. They say themselves, this is a 0:09:53.760 --> 0:09:56.600 thought experiment. There was something kind of gripping about the 0:09:56.600 --> 0:09:58.520 fact that it was claiming to be a memo written 0:09:58.559 --> 0:10:01.160 in twenty twenty eight. The claim was you would have 0:10:01.240 --> 0:10:03.520 an extraordinary amount of change in business models in a 0:10:03.600 --> 0:10:06.880 very short period of time. I assume you would say, 0:10:07.440 --> 0:10:12.280 given real world frictions and just the way things tend 0:10:12.360 --> 0:10:15.880 to happen, that a two year timeframe is very unrealistic. 0:10:16.160 --> 0:10:18.840 But there was another basic assumption built into that that 0:10:19.240 --> 0:10:21.320 the change that we will most immediately see and will 0:10:21.360 --> 0:10:24.480 have the biggest impact will be simply to replace labor, 0:10:24.960 --> 0:10:27.680 not to augment it, and that to the extent that 0:10:27.760 --> 0:10:31.559 it's creating other stuff that's going to be far outweighed 0:10:32.080 --> 0:10:34.960 by the job destruction. Even if you don't accept the 0:10:35.000 --> 0:10:38.000 time frame, do you think there is an emphasis on 0:10:38.559 --> 0:10:41.920 replacement relative to augmentation or is it a gent ki 0:10:42.040 --> 0:10:43.400 really both things. 0:10:43.720 --> 0:10:48.280 I don't know. It's an open question, but my bet 0:10:48.440 --> 0:10:52.040 would be on the Cititrini side, not on the timeframe, 0:10:52.720 --> 0:10:56.480 but on the path that we are following. There is 0:10:56.679 --> 0:11:00.960 so little that these companies are doing in order to 0:11:01.080 --> 0:11:05.800 understand what work humans do and try to be useful 0:11:05.800 --> 0:11:09.280 to humans. The whole agenda of all of the leading 0:11:09.320 --> 0:11:11.720 companies in the United States and now joined by deep 0:11:11.760 --> 0:11:17.520 Seek in China is agi artificial general intelligence. That is 0:11:17.559 --> 0:11:21.000 a banner for saying these models are going to do 0:11:21.080 --> 0:11:24.600 everything better than humans, which of course then leads to 0:11:24.640 --> 0:11:28.040 the corollary that a lot of companies should just throw 0:11:28.080 --> 0:11:30.920 away their humans and use these companies. That is an 0:11:30.960 --> 0:11:35.200 automation agenda that is exactly what Catrini banked on. Now, 0:11:35.520 --> 0:11:39.720 they then made a number of other assumptions and steps 0:11:39.760 --> 0:11:42.560 about how that would work out, what its consequences would be, 0:11:42.559 --> 0:11:44.760 how quickly those would be. Those I don't agree with, 0:11:45.440 --> 0:11:49.120 but credit to them. They said this was a scenario. 0:11:49.200 --> 0:11:51.839 They weren't even making a prediction. I don't know why 0:11:51.880 --> 0:11:55.040 the markets went haywire, given that there was no new 0:11:55.040 --> 0:11:57.640 information in there. Everything that was in the Sutrini report 0:11:57.679 --> 0:11:59.800 has been said, and they themselves said, there is no 0:12:00.120 --> 0:12:03.320 search here that's original. But you know, we are living 0:12:03.400 --> 0:12:08.480 in such fragile times in everything. You know, the valuation 0:12:08.600 --> 0:12:12.439 of these companies is all based on very fragile assumptions 0:12:12.480 --> 0:12:14.240 about what they're going to be able to achieve in 0:12:14.280 --> 0:12:17.360 the future. If you look at the amount that they're 0:12:17.400 --> 0:12:21.000 spending and their valuations, this can only be justify if 0:12:21.000 --> 0:12:25.160 they make something like a trillion dollar revenues in the 0:12:25.200 --> 0:12:28.559 foreseeable future as an AI industry. I mean, that's just incredible. 0:12:28.679 --> 0:12:30.920 How are you going to get to a trillion dollar there? 0:12:31.040 --> 0:12:33.480 They're hardly struggling to make a couple of billion dollars 0:12:33.559 --> 0:12:36.800 right now as a whole industry. So there is something, 0:12:37.520 --> 0:12:39.920 you know, glass in this house. 0:12:41.040 --> 0:12:42.720 Okay, I have to say that I'm slightly depressed by 0:12:42.720 --> 0:12:44.240 that answer because I thought you were going to push 0:12:44.280 --> 0:12:49.040 back more heavily against the Citrey assessment. Though I absolutely 0:12:49.080 --> 0:12:51.640 know that you're I know you have your concerned about 0:12:52.440 --> 0:12:55.280 our capacity to to cape with this, but I thought 0:12:55.360 --> 0:12:57.199 I let. 0:12:57.160 --> 0:12:59.040 Me give me out of the pushback as well. I mean, 0:12:59.080 --> 0:13:02.800 that's that's the point I want to make throughout. There 0:13:02.920 --> 0:13:06.959 is the potential to use AI not for automation. That's 0:13:06.960 --> 0:13:11.040 what I keep emphasizing. But I also want to push 0:13:11.240 --> 0:13:15.760 very hard against the assumption that either we are going 0:13:15.800 --> 0:13:17.920 there already, No we're not, or that we can get 0:13:17.920 --> 0:13:20.960 there automatically. No, we cannot. Because all of these companies 0:13:21.160 --> 0:13:25.360 have this business model or just let's repoliticplace all the workers. 0:13:25.640 --> 0:13:29.480 They haven't even put into their calculations much of a 0:13:30.480 --> 0:13:34.040 revenue stream that they can get from complementing workers, because 0:13:34.080 --> 0:13:36.920 that's just a very difficult thing to monetize. So I 0:13:36.920 --> 0:13:41.440 think that's where our wisdom gap is. We are not 0:13:41.760 --> 0:13:45.240 even wisely thinking about what we should be doing with 0:13:45.840 --> 0:13:49.760 these very capable models, and the industry is going in 0:13:49.800 --> 0:13:50.720 its own direction. 0:13:51.480 --> 0:13:54.320 You have just done a paper with two of your 0:13:54.360 --> 0:13:57.320 colleagues for Brookings that is trying to give some concrete 0:13:57.320 --> 0:14:00.400 advice to policymakers in this area to answer specifically that question. 0:14:00.720 --> 0:14:01.960 I do want to get to that. I just want 0:14:02.000 --> 0:14:05.160 to quickly one of the things, just to make sure 0:14:05.280 --> 0:14:07.160 coming out of this conversation. One of the things that 0:14:07.160 --> 0:14:10.160 we see a lot in this is, and particularly if 0:14:10.160 --> 0:14:14.320 you look at the research studies in this area, they 0:14:14.440 --> 0:14:17.440 tend to look at the range of occupations and talk 0:14:17.480 --> 0:14:21.200 about their degree of exposure quote unquote to AI. And 0:14:21.240 --> 0:14:23.640 there's a whole range, I think in your original assessment, 0:14:23.680 --> 0:14:25.320 and a lot of people use this. They sort of 0:14:25.360 --> 0:14:28.360 thought it was about twenty percent of occupations, and obviously 0:14:28.440 --> 0:14:31.200 some people have high numbers. If one's thinking about different 0:14:31.280 --> 0:14:34.120 kinds of AI and the potential of different of AI, 0:14:34.680 --> 0:14:37.480 how should we read those? Are they just going on 0:14:37.520 --> 0:14:39.880 your the way that businesses is looking at it now, 0:14:39.920 --> 0:14:42.040 as you pointed out as very much as a labor 0:14:42.080 --> 0:14:46.440 replacement technology, should we see that as exposure to replacement 0:14:46.560 --> 0:14:48.560 or should we see it something more sophisticated? 0:14:48.800 --> 0:14:51.120 Great set of questions. There are really three questions you're 0:14:51.120 --> 0:14:53.720 asking here, Stephanie. Let me answer each one of them 0:14:53.760 --> 0:14:56.920 in turn. One is what does this AI exposure mean? 0:14:57.840 --> 0:15:02.560 And in general it is an ill defined concept because 0:15:02.600 --> 0:15:05.160 you could be exposed to AI because you can lose 0:15:05.200 --> 0:15:08.560 your job with AI, or you could be exposed to 0:15:08.600 --> 0:15:12.160 AI because you could use AI to increase your contribution 0:15:12.240 --> 0:15:12.840 to your job. 0:15:13.000 --> 0:15:15.560 And we see that in the company's exposure as well. 0:15:15.600 --> 0:15:18.400 And investors can't decide the difference between those two either. 0:15:19.200 --> 0:15:24.200 That's why they fluctuate between devaluing software companies and giving 0:15:24.240 --> 0:15:27.320 them a huge boost. So that's the first problem with 0:15:27.400 --> 0:15:30.600 AI exposure. So when I wrote my paper, I took 0:15:31.080 --> 0:15:33.440 a position similar to the Citrenia report, and I said, 0:15:33.520 --> 0:15:35.800 right now, we're going towards automation, so let me focus 0:15:35.840 --> 0:15:42.440 on that second. Where does that twenty percent number come from? So, 0:15:42.600 --> 0:15:45.960 roughly speaking, think of it this way right now, and 0:15:46.040 --> 0:15:51.000 I think in the near future, AI is pretty useless 0:15:51.040 --> 0:15:55.440 in jobs that involve a huge amount of interaction with 0:15:55.520 --> 0:16:00.960 the physical world construction, custodial work, manufacturing work, work that 0:16:01.040 --> 0:16:07.320 involves home care, airdressing, hairdressing. The reason being that we 0:16:07.440 --> 0:16:12.200 are very far behind in robotics, but also AI models 0:16:12.240 --> 0:16:16.240 themselves don't have a good conceptual understanding of spatial causal 0:16:16.360 --> 0:16:20.920 relations that even if we had fantastically flexible robots that 0:16:21.000 --> 0:16:25.720 could cut your hair or hold your hand. AI models 0:16:25.720 --> 0:16:30.480 would continuously make mistakes about spatial coosal relations and those 0:16:30.560 --> 0:16:35.040 unreliabilities would end up breaking your neck. So let's eliminate 0:16:35.120 --> 0:16:39.280 those jobs. I've also eliminated, again based on other people's coding, 0:16:39.520 --> 0:16:44.080 any jobs that include a high degree of judgment. So 0:16:44.160 --> 0:16:51.360 we wouldn't want AI to run air traffic control. So Stephanie, 0:16:51.400 --> 0:16:54.880 think about it yourself. If the Manchester Airport said from 0:16:54.920 --> 0:16:57.200 now on, we're not going to have any air traffic controlers. 0:16:57.200 --> 0:16:59.320 Everything's gonna be done by AI. It might hallucinate, it 0:16:59.360 --> 0:17:02.200 might make some mistakes, but that's fine. It's cost savings. 0:17:02.280 --> 0:17:05.520 Would you fly to Manchester Airport? So we don't want that. 0:17:06.280 --> 0:17:09.520 So those jobs are out, and any job that involves 0:17:09.520 --> 0:17:13.200 a high degree of social interaction is out as well. 0:17:13.240 --> 0:17:17.639 So that leaves essentially a range of office cognitive jobs. 0:17:17.840 --> 0:17:20.159 So that's where the twenty percent comes from. Now what 0:17:20.240 --> 0:17:24.120 about companies. So companies are indeed going after those jobs. 0:17:24.119 --> 0:17:27.240 They're going after it jobs, they're going after back office jobs. 0:17:27.560 --> 0:17:30.800 But there are several new papers that have come out 0:17:30.920 --> 0:17:33.280 over the last few months and they all find the 0:17:33.280 --> 0:17:37.040 same things. The companies are talking a big game about AI. 0:17:37.240 --> 0:17:39.840 They say, oh, we have a lot of AI being used, 0:17:40.160 --> 0:17:42.640 but it has so far zero impact on the companies, 0:17:43.240 --> 0:17:48.080 zero impact on employment, zero impact on productivity because it's 0:17:48.160 --> 0:17:53.080 actually like just other technologies, it's spread slowly. That's us 0:17:53.160 --> 0:17:57.080 the basis of my numbers. And it's very difficult to 0:17:57.200 --> 0:18:00.480 integrate AI into what those companies do with the big 0:18:00.600 --> 0:18:03.560 organizational change. And I think when actually push comes to shove, 0:18:04.359 --> 0:18:07.040 when they try the organizational change, but they're going to 0:18:07.119 --> 0:18:11.479 realize that you cannot really replace it security people with AI. 0:18:11.520 --> 0:18:15.040 You need to use it security people together with AI, 0:18:15.080 --> 0:18:17.919 and that might actually give us a boost towards more human, complementary, 0:18:17.960 --> 0:18:20.200 more pro worker AI. But we're not there yet because 0:18:20.280 --> 0:18:23.080 they're not trying to do that in big scale yet. Now, 0:18:23.359 --> 0:18:27.200 of course code that's a big advance. Will that change 0:18:27.200 --> 0:18:30.440 things in twenty twenty six, I don't know. By twenty 0:18:30.480 --> 0:18:33.200 twenty seven, I'm sure there will be more companies that 0:18:33.640 --> 0:18:36.040 have attempted to do things, and perhaps we'll have a 0:18:36.119 --> 0:18:38.080 rude awakening and this is not going to work in 0:18:38.080 --> 0:18:40.080 the way that we're trying to do it. Perhaps we'll 0:18:40.080 --> 0:18:43.080 find a new direction. But this is where both policy 0:18:43.119 --> 0:18:45.119 and public debate are really important. 0:18:45.480 --> 0:18:47.479 To your point, I think Goldman Sachs added up if 0:18:47.480 --> 0:18:49.879 you just listen to all the earnings sort of calls 0:18:49.960 --> 0:18:52.720 that companies are doing and chief executives are giving. Just 0:18:52.760 --> 0:18:54.840 to your point about all these companies claiming I think 0:18:54.880 --> 0:18:57.919 the average productivity growth that they're claiming is thirty two percent, 0:18:58.520 --> 0:19:06.400 but it's not necessary evident in any of the numbers. 0:19:13.840 --> 0:19:16.680 Let's get onto what policy makers could do about it, 0:19:17.160 --> 0:19:20.920 because that's something that governments everywhere are obviously very focused on. 0:19:21.400 --> 0:19:24.600 And I noticed that you had recently done this report 0:19:24.600 --> 0:19:27.760 for Brookings. I think about a framework for thinking about 0:19:27.840 --> 0:19:31.520 pro worker AI. What are the main sort of policy 0:19:31.560 --> 0:19:33.639 areas that you would like people to focus on for that. 0:19:33.960 --> 0:19:36.800 First of all, just two points I want to make 0:19:37.560 --> 0:19:40.040 before I talk about policy. The first one is just 0:19:40.080 --> 0:19:42.720 to clarify that by pro worker AI, I mean exactly 0:19:42.720 --> 0:19:44.359 the same thing that I was just talking about a 0:19:44.440 --> 0:19:48.720 second ago. Human complementary AIAI that helps workers do more, 0:19:49.200 --> 0:19:53.240 help workers become more expert in their jobs, perform new tasks, 0:19:53.640 --> 0:19:57.240 have better information for problem solving, travel shooting, judgment, and 0:19:57.280 --> 0:19:59.880 so on. That's what we're talking about with pro worker 0:20:00.200 --> 0:20:02.000 and not just for office workers. We have a lot 0:20:02.000 --> 0:20:05.720 of examples in the paper showing how manual workers can 0:20:05.760 --> 0:20:10.960 benefit from AI. It cannot replace manual workers, but electricians, plumbers, 0:20:11.119 --> 0:20:14.760 nurses can hugely benefit from having the right kind of 0:20:14.800 --> 0:20:17.280 AI assistant. But it has to be the right kind 0:20:17.280 --> 0:20:19.760 of aassistant. It's not going to be chut GBT. So 0:20:19.800 --> 0:20:22.040 that's the first point. The second point is that my 0:20:22.160 --> 0:20:26.840 belief is that as important as policy actually is what 0:20:26.880 --> 0:20:31.640 we're doing right now, Stephanie, is the public debate. Right now, 0:20:31.680 --> 0:20:37.880 we have delegated the future of this very very important technology. 0:20:38.119 --> 0:20:40.719 Some would argue, therefore the future of humanity to a 0:20:40.760 --> 0:20:44.760 handful of people who have no feedback from society, who 0:20:44.800 --> 0:20:49.080 have no accountability to society, and right now society is confused. 0:20:50.359 --> 0:20:54.200 So on our current path, this handful of people will 0:20:54.200 --> 0:20:57.840 decide what the future of AI is. And the best 0:20:57.880 --> 0:21:02.480 way to counter that is to have a vision that's 0:21:02.520 --> 0:21:06.240 different and hopefully better for society. And that's what I 0:21:06.280 --> 0:21:09.000 hope the pro worker AI vision is. So the more 0:21:09.040 --> 0:21:12.640 people talk about that, the more the public pressure will grow, 0:21:12.680 --> 0:21:15.120 and the more of an alternative there will be. Look, 0:21:15.720 --> 0:21:21.560 my understanding from my limited experience is that anthropic Google 0:21:21.880 --> 0:21:26.200 Open AI are filled with people who are very well meaning. 0:21:26.680 --> 0:21:30.439 If they thought that there is a socially beneficial and 0:21:30.520 --> 0:21:34.640 still technically exciting area of AI, they would be much 0:21:34.640 --> 0:21:39.520 more likely to take the plunge in that direction. It's 0:21:39.600 --> 0:21:42.640 just that we're not offering them an alternative, and society 0:21:42.720 --> 0:21:46.760 is not pushing back against some Outlin and his ilk's vision. 0:21:46.840 --> 0:21:49.239 So that's the point. Policy, in my view, is a 0:21:49.280 --> 0:21:53.639 supporting set of instruments. It can remove the stortions that 0:21:53.760 --> 0:21:58.800 exist that solidify the existing system, and it can give 0:21:58.840 --> 0:22:02.800 a notch to people, as policy has done in the past, 0:22:03.040 --> 0:22:06.840 to try new things. So on the first bucket, there 0:22:06.880 --> 0:22:12.320 are many problems in our current system that would make 0:22:13.480 --> 0:22:17.200 a redirection of AI in a pro worker direction more difficult. 0:22:17.320 --> 0:22:19.159 I would single out two of them, but there are 0:22:19.200 --> 0:22:22.560 is more. The first one is that our tax code. 0:22:22.880 --> 0:22:24.840 That's true in the UK, that's true in the US. 0:22:25.320 --> 0:22:30.280 Our tax code encourages firms to replace workers because we 0:22:30.400 --> 0:22:34.880 tax capital essentially at zero percent, labor twenty five to 0:22:34.920 --> 0:22:37.480 thirty percent, especially in the US once you had the 0:22:37.520 --> 0:22:40.800 healthcare costs and all the payroll taxes and everything. So 0:22:40.840 --> 0:22:45.959 that's a massive subsidy to capital that would make firms 0:22:46.040 --> 0:22:50.280 adopt automation even if automation wasn't better than humans, because 0:22:50.280 --> 0:22:55.080 they're getting this subsidy. Second, we know from historical evidence 0:22:55.119 --> 0:22:58.840 and current evidence that new things are done by new firms. 0:23:00.160 --> 0:23:05.000 Competition is really important. The tech industry has become one 0:23:05.000 --> 0:23:12.280 of the least competitive industries in history, and moreover, business 0:23:12.320 --> 0:23:16.240 models that are new and different are likely to get crushed. 0:23:16.320 --> 0:23:23.720 So encouraging more competition via antitrust by enabling new companies 0:23:23.800 --> 0:23:25.560 to enter and try new things, I think that's a 0:23:25.640 --> 0:23:27.400 very important part of way. Now there is a lot 0:23:27.400 --> 0:23:30.920 of energy in Silicon Valley, but it's all these startups 0:23:30.920 --> 0:23:33.240 that try to do exactly what open Ai and Anthropic 0:23:33.359 --> 0:23:35.560 and Google do so that they can be both by them. 0:23:35.560 --> 0:23:37.680 So that's not the kind of competition I'm talking about. 0:23:37.880 --> 0:23:41.639 And then in terms of nudging us to do new things, 0:23:42.240 --> 0:23:45.640 the government is horrible, in my opinion, at being an entrepreneur. 0:23:45.720 --> 0:23:48.240 It cannot be a venture capitalist, it cannot be an entrepreneur, 0:23:48.280 --> 0:23:52.640 it cannot be an innovator, but it has great potential 0:23:53.320 --> 0:23:58.160 to be an aspiring leader. We have had so many 0:23:58.240 --> 0:24:01.480 examples where a small amount of money from the government 0:24:02.440 --> 0:24:08.080 has kickstarted industries in nanotechnology, in the internet, in robotics, 0:24:08.080 --> 0:24:11.439 it was the Robotics Challenge, a million dollar challenge that 0:24:11.560 --> 0:24:14.480 really focused people's attention to get robots that could actually 0:24:14.480 --> 0:24:17.520