WEBVTT - Rot Economics with MIT's Daron Acemoglu 0:00:02.160 --> 0:00:03.400 Also media. 0:00:04.640 --> 0:00:07.120 Hello and welcome to Better Offline. I'm your host ed 0:00:07.240 --> 0:00:21.680 ze tron. Not really going to dawdle too much on 0:00:21.680 --> 0:00:24.000 the intro today, I'm too excited about the interview you're 0:00:24.000 --> 0:00:27.440 about to hear. Today's guest. There's a globally known economist, 0:00:27.680 --> 0:00:30.440 the author of wy Nations, Fall and Power and Progress. 0:00:30.880 --> 0:00:35.239 Am I tas the rhone as sema glue. Okay, So 0:00:36.120 --> 0:00:40.800 a term that you've popularized, not necessarily invented, is creative destruction. 0:00:41.080 --> 0:00:43.040 Do you mind explaining it for the listeners? 0:00:43.400 --> 0:00:45.760 Oh? Yeah, I mean I definitely did not invent it, 0:00:45.800 --> 0:00:49.320 and I think many other people deserve much more credit 0:00:49.400 --> 0:00:52.839 for inventing it and making it work. It's this idea 0:00:53.120 --> 0:00:57.840 that goes back to Joseph Schumpeter, famous Austrian economist who 0:00:57.920 --> 0:01:00.840 spent most of his career or the most important part 0:01:00.840 --> 0:01:06.440 of his career, at Harvard, who emphasized that in capitalist growth, 0:01:07.959 --> 0:01:13.880 you will have new firms taking market away and destroying 0:01:13.920 --> 0:01:19.600 old firms, and as a corollary of that, new technologies 0:01:20.160 --> 0:01:24.880 taking market share away and driving out old technologies. And 0:01:25.360 --> 0:01:30.440 he understood this was a difficult and tumultuous process, but 0:01:30.600 --> 0:01:35.000 also believe that that was the essence of sort of 0:01:35.000 --> 0:01:40.039 capitalist growth. So it's one of these things that is 0:01:40.080 --> 0:01:44.360 a fact of life in a market process, but different 0:01:44.520 --> 0:01:49.800 types of social, economic, and political reactions to it are natural, 0:01:50.000 --> 0:01:53.280 and how you react to it is going to have 0:01:53.680 --> 0:01:56.200 various effects both on growth, what type of growth, and 0:01:56.240 --> 0:01:57.680 its distributional effects. 0:01:58.560 --> 0:02:01.560 Right, So something I've read about and spoken about a 0:02:01.600 --> 0:02:03.560 lot as well as like this idea of the rot economy, 0:02:03.600 --> 0:02:06.600 which is the kind of growth has overtaken most of 0:02:06.640 --> 0:02:10.240 the modern markets. And I'd argue, at least in the 0:02:10.400 --> 0:02:15.800 creative destruction field, tech has stopped really innovative. It doesn't 0:02:15.840 --> 0:02:19.200 feel like they're creating things to create new jobs, to 0:02:19.200 --> 0:02:22.919 create new markets. I'm wondering how you feel about looking 0:02:22.919 --> 0:02:24.240 at the general tech industry. 0:02:24.960 --> 0:02:28.880 I am a critic of the tech industry, and I 0:02:28.919 --> 0:02:33.799 have become so over the last decade or so. And 0:02:34.200 --> 0:02:38.160 my problem with the tech industry is not its dynamism. 0:02:38.360 --> 0:02:41.840 I applaud that. It's not its risk taking. I applaud that, 0:02:42.720 --> 0:02:46.520 And it's not the drive towards economic growth, which I 0:02:46.560 --> 0:02:53.040 think is also generally desirable. But it is the direction 0:02:53.880 --> 0:02:57.760 of research and technologies that the tech industry has focused on. 0:02:58.880 --> 0:03:03.760 Both because of idea logical reasons and because of a 0:03:03.840 --> 0:03:08.040 particular business model that they developed, and I think both 0:03:08.080 --> 0:03:12.840 of those have pushed us towards technologies that I see 0:03:12.840 --> 0:03:17.640 as socially less desirable in some cases actually undesirable, and 0:03:17.680 --> 0:03:21.000 as a result, we're actually getting growth without as much 0:03:21.400 --> 0:03:25.600 social benefits. And let me try to just make one 0:03:25.720 --> 0:03:29.680 very simple point, which everybody nods when you say that, 0:03:29.760 --> 0:03:32.720 but it's still as important to put in the conversation. 0:03:33.240 --> 0:03:38.200 Economic output, as measured by statistical agencies such as gross 0:03:38.200 --> 0:03:43.280 domestic product, does not have any welfare element in it. 0:03:44.320 --> 0:03:48.440 So if I find a way of hacking into your 0:03:48.480 --> 0:03:52.360 computer spending one thousand dollars, and you find a way 0:03:52.360 --> 0:03:57.400 of defending against me spending two thousand dollars, that will 0:03:57.560 --> 0:04:00.960 increase GDP by three thousand dollars. And I think even 0:04:01.000 --> 0:04:04.080 the most demented person wouldn't say that that's a social improvement. 0:04:04.440 --> 0:04:05.880 Yeah, I think you made a point like this the 0:04:05.920 --> 0:04:08.400 Golden Zachs, where it's about like you could make a 0:04:08.480 --> 0:04:11.280 trillion dollars if you did deep fakes in a certain 0:04:11.320 --> 0:04:12.520 manner exactly. 0:04:13.440 --> 0:04:18.039 So, therefore, new products that increase GDP may have socially 0:04:18.120 --> 0:04:25.040 undesirable consequences. That wasn't part of the original Schumpeter point, 0:04:25.440 --> 0:04:27.680 and it's not something that I would worry about when 0:04:27.720 --> 0:04:31.560 I'm talking to people in Mexico who are trying to 0:04:31.560 --> 0:04:34.040 get the economy going, But it is a very important 0:04:34.080 --> 0:04:35.479 concern when it comes to new tech. 0:04:35.720 --> 0:04:37.520 When do you think this shift happened? You say, the 0:04:37.600 --> 0:04:40.240 last decade? What was it that kind of changed for you? 0:04:40.880 --> 0:04:45.000 Well, I think it was probably a gradual process. But 0:04:45.400 --> 0:04:50.400 the tech sector initially was very heavy on hardware with 0:04:50.560 --> 0:04:55.480 some software elements, right, And when that started changing and 0:04:55.560 --> 0:04:59.640 the entire field became software, I think the possibilities for 0:04:59.680 --> 0:05:03.360 different diferent types of technologies to go in very different 0:05:03.440 --> 0:05:09.919 social directions also multiplied. Right. So money today in Vidia 0:05:09.920 --> 0:05:13.960 being an exception, is not made by hardware, And even 0:05:13.960 --> 0:05:16.520 in Nvidia, I think a lot of the innovation is with. 0:05:17.080 --> 0:05:20.359 Software, particularly with Kuda being able to do stuff with 0:05:20.400 --> 0:05:21.640 GPS exactly. 0:05:21.680 --> 0:05:25.760 Yeah. But when you are also doing software, you have 0:05:26.400 --> 0:05:29.440 ways in which that software becomes an information control tool, 0:05:29.680 --> 0:05:32.600 a monitoring tool, or surveillance tool. It becomes a way 0:05:32.640 --> 0:05:36.359 of automating work in various different ways. It can become 0:05:36.440 --> 0:05:40.159 a manipulative tool, and it can also create lots of 0:05:40.240 --> 0:05:42.800 new products, some of them very beneficial, but some of 0:05:42.839 --> 0:05:46.680 them very addictive and conducive to mental health problems. So 0:05:46.680 --> 0:05:52.120 I think software sort of expands the capabilities, but together 0:05:52.160 --> 0:05:56.200 with the capabilities, you also have expanded set of distortionary 0:05:56.240 --> 0:05:57.840 or manipulative things that you can do. 0:05:59.040 --> 0:06:02.159 Right, you mentioned what kind of dynamism element of how 0:06:02.240 --> 0:06:04.760 tech is working in the growth? How if I agree 0:06:05.360 --> 0:06:08.080 that tech is in a dynistic state, it almost feels 0:06:08.120 --> 0:06:12.040 like it's been spinning its wheels for the last few years. Crypto, metaverse, 0:06:12.080 --> 0:06:13.920 all of this stuff. It doesn't feel like new things 0:06:13.920 --> 0:06:14.520 are happening. 0:06:15.279 --> 0:06:17.400 Well, they are new things, it's just that I think 0:06:17.440 --> 0:06:22.040 you are saying what I just said in a different way, 0:06:22.040 --> 0:06:24.600 and it might be that your way is better. They're 0:06:24.600 --> 0:06:27.400 not super socially valued, but they are new products. So 0:06:27.440 --> 0:06:32.040 you would say that metaverse or virtual worlds are a 0:06:32.040 --> 0:06:35.200 new product by any category. It's not just not something 0:06:35.240 --> 0:06:37.240 that's going to make humanity better if it might make 0:06:37.320 --> 0:06:41.120 humanity worse by alienating them more or isolating them more 0:06:41.160 --> 0:06:44.120 from their social milieu. You know. And there is another 0:06:44.160 --> 0:06:49.120 element of what's going on here, which I'll comment on 0:06:49.200 --> 0:06:52.200 but again it doesn't contradict that they are generating new products. 0:06:52.240 --> 0:06:55.320 A lot of new technologies and new ideas that are 0:06:55.360 --> 0:06:59.680 being invented in tech are not being implemented, and part 0:06:59.760 --> 0:07:03.880 of the reason for that is the competitive environment. Google, Facebook, Microsoft, 0:07:03.920 --> 0:07:07.880 Amazon are all buying up on Facebook, are all buying 0:07:07.960 --> 0:07:11.800 up a lot of competitors and not even sometimes using 0:07:11.840 --> 0:07:15.160 their technology. So that's the consolidated structure. So the invention 0:07:15.280 --> 0:07:18.760 is there, but the invention is not translating into implementation. Now, 0:07:18.920 --> 0:07:21.440 don't get me wrong, some of that invention may also 0:07:21.480 --> 0:07:24.360 go in the wrong way, so a better version of 0:07:25.400 --> 0:07:28.520 you know, TikTok may not be a great thing either, 0:07:28.800 --> 0:07:34.240 But there is that consolidated, concentrated market structure that is 0:07:34.400 --> 0:07:36.760 also changing what gets implemented. 0:07:37.040 --> 0:07:39.120 The thing I'll push back on is they are making 0:07:39.240 --> 0:07:43.000 new things, but it feels like the ones I mentioned Crypto, 0:07:43.320 --> 0:07:48.600 Metaverse and now Generative AI. They're not producing actual products 0:07:48.600 --> 0:07:50.280 at the end. It's not so much that we couldn't 0:07:50.320 --> 0:07:52.560 live in a virtual world or that digital money wonn't 0:07:52.560 --> 0:07:56.600 be usedful, but it's more that the actual output from 0:07:56.640 --> 0:08:00.000 the companies is not translating into meaningful products. 0:08:01.000 --> 0:08:03.640 Well, again, this is a question of what we are measuring, 0:08:03.680 --> 0:08:05.480 and whether what we're measuring is the right thing, and 0:08:05.520 --> 0:08:08.720 whether it's really welfare relevant. But if I create a 0:08:08.760 --> 0:08:14.240 metaverse and you're willing to pay a million dollars for it, 0:08:14.960 --> 0:08:17.520 that will increase GDP by a million dollars. So that's 0:08:17.560 --> 0:08:19.920 a new service, right. So a lot of things we 0:08:20.000 --> 0:08:24.000 consume today are services. It's not something produced like a 0:08:24.040 --> 0:08:28.200 T shirt or a car. They are based on digital services. Now, 0:08:28.200 --> 0:08:30.480 of course, to produce those digital services, we're actually using 0:08:30.480 --> 0:08:33.640 real resources such as energy. But what I actually buy 0:08:33.640 --> 0:08:35.520 from you maybe, or what you buy from you maybe 0:08:35.559 --> 0:08:39.240 just a digital service. Now, some digital services are extremely useful, 0:08:39.760 --> 0:08:41.440 and some of them are useless. Some of them may 0:08:41.480 --> 0:08:42.120 be bad for. 0:08:42.040 --> 0:08:47.000 You, right. I think my point is more they're not 0:08:47.120 --> 0:08:50.400 making particularly useful Selvis. They're doing well monetizing the things 0:08:50.400 --> 0:08:52.640 they've had for years. But it kind of reminds me 0:08:52.679 --> 0:08:55.760 of something he wrote in twenty nineteen where you were 0:08:55.760 --> 0:09:00.680 talking about automation and effect on growth, but also we 0:09:00.760 --> 0:09:03.520 may have run out of ideas for generating new high productivity, 0:09:03.600 --> 0:09:06.600 labor intensive tasks. Do you think we're approaching that point? 0:09:07.240 --> 0:09:09.160 Thank you for raising that, ed So. I think that 0:09:09.240 --> 0:09:12.240 is a very very important part of my thinking. I 0:09:12.240 --> 0:09:16.599 would say that the tech sector is not producing sufficient 0:09:16.960 --> 0:09:21.360 new tasks for workers to use their skills and to 0:09:21.400 --> 0:09:25.000 expand their capabilities, and firms perhaps are not demanding and 0:09:25.040 --> 0:09:28.960 implementing enough of them. But it is not, according to me, 0:09:30.600 --> 0:09:33.480 because we're running out of possibilities. It's just that we 0:09:33.520 --> 0:09:36.720 haven't focused on those. And that's where the ideology reference 0:09:36.800 --> 0:09:39.840 earlier on that I made comes in. You know, I 0:09:39.880 --> 0:09:46.559 think the software industry would have done somewhat more productive 0:09:46.640 --> 0:09:52.679 things if it did not become too focused on replacing 0:09:52.760 --> 0:09:57.920 humans having machines as humans overlords, which today has of 0:09:57.960 --> 0:10:00.680 course reached its apex with the craze on AGI. 0:10:05.880 --> 0:10:07.720 But here's the thing. I get that that might be 0:10:07.760 --> 0:10:11.640 what they're pushing toward, But generative AI isn't even automation 0:10:11.960 --> 0:10:13.520 at this point. A lot of what you've written about 0:10:13.520 --> 0:10:16.520 AI is correctly like the effects of if we all 0:10:16.600 --> 0:10:19.640 make these tasks, but it feels that they aren't even 0:10:19.800 --> 0:10:21.800 successful automating anything. 0:10:21.520 --> 0:10:25.640 Absolutely one percent. Thank you for saying that. My take 0:10:25.800 --> 0:10:30.800 is that generative AI is actually an informational tool, right, 0:10:31.160 --> 0:10:38.800 so you should use generative AI as a way of generating, filtering, summarizing, finding, 0:10:39.400 --> 0:10:44.640 checking information. That's actually what it's good at. If you 0:10:44.800 --> 0:10:47.400 try to use it for other purposes, sometimes you can 0:10:47.440 --> 0:10:49.199 get away with it, but it won't be very good 0:10:49.280 --> 0:10:52.000 at that. So you can try to automate a lot 0:10:52.040 --> 0:10:56.760 of warehouse tasks today by using the current crop of robots. 0:10:57.440 --> 0:11:00.000 People don't do that because they are not good at it. 0:11:01.200 --> 0:11:03.960 If you did it, costs would go up, people would 0:11:04.000 --> 0:11:06.920 lose their jobs, Delays would pile up. But you can 0:11:07.080 --> 0:11:09.520 do it. It's the same thing with generative AI. Even 0:11:09.520 --> 0:11:12.600 though it's not an automation tool. Automation wouldn't be its 0:11:12.640 --> 0:11:17.080 best use, especially given the current unreliability, even though there 0:11:17.160 --> 0:11:19.480 is something else that we could do better with it. 0:11:20.200 --> 0:11:21.960 I think many people are going to use it for 0:11:22.000 --> 0:11:25.319 automation because that's the vibe. That's what companies are being told. 0:11:25.720 --> 0:11:28.760 You know, if you talk to business leaders today, everybody's 0:11:28.800 --> 0:11:32.800 asking them, financial journalists, their shareholders, and their friends, where 0:11:32.840 --> 0:11:36.120 are you with the AI investment? So that's the hype, 0:11:37.160 --> 0:11:39.480 and then people are going to rush into use AI 0:11:39.600 --> 0:11:41.600 to implement AI even when they don't know what to 0:11:41.640 --> 0:11:44.720 do with it, and automation will often appeal to them. 0:11:44.760 --> 0:11:46.800 Because it's like the easiest thing to do. It's the 0:11:46.840 --> 0:11:50.680 thing that they may have experienced from other technologies, and 0:11:50.720 --> 0:11:52.440 it's the thing that some people are telling them that 0:11:52.440 --> 0:11:55.840 that's what they should do. There are companies, integrators' websites 0:11:55.880 --> 0:11:58.920 devoted to automation AI, even though it wouldn't be very 0:11:58.920 --> 0:12:02.040 good at it. I mean again, it could automate some tasks. 0:12:02.360 --> 0:12:06.760 You could have more of your customer service done without people. 0:12:07.600 --> 0:12:10.320 Right, even then, that feels like a stretch of what 0:12:10.400 --> 0:12:14.080 automation means because sure, the customer service example, and I 0:12:14.120 --> 0:12:16.760 think you may have raised this point as well, how 0:12:16.760 --> 0:12:19.320 does it get better? How do you measure better in 0:12:19.320 --> 0:12:22.240 the case of customer service? But even then, it's automation 0:12:22.440 --> 0:12:24.680 only in so much as you can trust it, And 0:12:24.720 --> 0:12:28.079 it feels like these core issues of hallucination almost kill 0:12:28.920 --> 0:12:31.000 the concept of automation with generative AI. 0:12:32.320 --> 0:12:35.600 So here's a good use case for automation of customer service, 0:12:35.760 --> 0:12:38.720 which is you call your bank and you enter some 0:12:38.760 --> 0:12:42.400 password and they tell you your balance. That's perfect. You 0:12:42.440 --> 0:12:45.120 don't need a person there to tell you the balance, right, 0:12:45.440 --> 0:12:50.680 because the current technologies can faithfully take those numbers and 0:12:50.720 --> 0:12:53.280 communicate them to you. After the right security steps. 0:12:53.880 --> 0:12:56.320 Yeah, and it's not a generative answer because it's a 0:12:56.520 --> 0:12:57.400 number in a day abase. 0:12:57.480 --> 0:13:00.080 It's not a generative answer. So now put generitive A 0:13:00.280 --> 0:13:02.480 and there you're probably gonna get lots of incorrect answers. Yes, 0:13:02.960 --> 0:13:04.480 but some companies might still do it. 0:13:04.520 --> 0:13:07.400 Which is it just feels like a crazy time that 0:13:07.440 --> 0:13:10.320 you have companies shoving this through almost like the very 0:13:10.400 --> 0:13:14.200 much like a post Jack Welch situation where. 0:13:14.160 --> 0:13:16.600 Yes, exactly the Jack Welsh mindset. 0:13:20.480 --> 0:13:22.080 Do you think about the problem is that the people 0:13:22.160 --> 0:13:24.560 running these companies aren't really technologists. 0:13:25.520 --> 0:13:29.000 I don't know. Look, I think this is another branch 0:13:29.080 --> 0:13:35.839 of my work. But US businesses are often led by 0:13:35.960 --> 0:13:41.320 people who have been trained into thinking that their only 0:13:41.640 --> 0:13:45.800 priority should be increasing short term shareholder value, right, and 0:13:46.520 --> 0:13:49.440 a very effective way of doing that is cut labor costs. 0:13:50.000 --> 0:13:53.920 But a that's not the right social objective. Even maximizing 0:13:53.960 --> 0:13:57.960 long term shareholder value is not the right objective. But 0:14:00.760 --> 0:14:06.640 even more fundamentally, cutting short run labor costs maybe an 0:14:06.640 --> 0:14:11.439 illusion and be associated with longer term problems. So if 0:14:11.480 --> 0:14:15.360 you have a company where your workers are skilled, talented, 0:14:15.400 --> 0:14:20.480 they are very useful for liaising with customers. Creating new services, products, innovation. 0:14:21.320 --> 0:14:23.240 You can in the short run cut your labor costs, 0:14:23.280 --> 0:14:25.360 but it would destroy you in the long run. I 0:14:25.360 --> 0:14:27.880 think many more companies are in this bucket than American 0:14:27.920 --> 0:14:29.080 business leaders realize. 0:14:29.200 --> 0:14:30.760 And it's funny you mentioned that. I've heard a lot 0:14:30.800 --> 0:14:33.400 from Google people. In particular. I get emails from Google 0:14:33.400 --> 0:14:35.240 people all the time because it did an episode about 0:14:35.240 --> 0:14:39.360 a particular guy, and it's funny. They all talk about 0:14:39.400 --> 0:14:42.200 the kind of brain drain of layoffs that you don't realize. 0:14:42.240 --> 0:14:45.160 It's not just the output you're losing, it's the person 0:14:45.160 --> 0:14:47.880 who knew how the staff worked and where the stuff was, 0:14:47.920 --> 0:14:49.920 and who built the staff and why the stuff is 0:14:49.960 --> 0:14:54.040 good or back. And it almost feels like American capitalism 0:14:54.120 --> 0:14:56.960 is dramatically disconnected from labor in general. 0:14:57.680 --> 0:15:00.400 Yeah. Absolutely so. Look, I mean, I think there is 0:15:00.520 --> 0:15:05.680 a tremendous amount of tacit knowledge that workers have which 0:15:05.720 --> 0:15:11.400 often gets unrecognized, and even bosses sometimes don't recognize that. 0:15:13.240 --> 0:15:16.960 So both French and British trade unions in the history 0:15:17.280 --> 0:15:21.560 experimented with these types of strikes where workers just follow 0:15:21.600 --> 0:15:25.080 the rules. They do exactly what the rule book says 0:15:25.080 --> 0:15:28.760 their responsibilities are, and it turns out to be quite 0:15:28.760 --> 0:15:32.320 disastrous for the company because most of what actually workers 0:15:32.360 --> 0:15:35.520 do is much more adaptive than just following the rule. 0:15:35.880 --> 0:15:37.680 Like kind of outsourcing risk. Almost. 0:15:38.200 --> 0:15:40.440 Yeah, it's just like, you know, the rule book says, 0:15:41.000 --> 0:15:43.200 you know, operate the machinery, but you know when to 0:15:43.280 --> 0:15:46.800 actually operate the machinery, not just operate the machinery. The 0:15:47.240 --> 0:15:52.160 sound yeah exactly. Yeah. So that's the kind of tacit 0:15:52.240 --> 0:15:55.920 knowledge that people acquire via training, why experienced by their 0:15:55.960 --> 0:16:00.280 social network, talking to friends. And if we don't value that, 0:16:00.320 --> 0:16:02.200 will lose that and it's going to be very difficult 0:16:02.240 --> 0:16:06.200 to replace it with what machines or information technologies do. 0:16:06.200 --> 0:16:07.720 Do you think we're in a bubble right now? 0:16:08.360 --> 0:16:09.160 Define a bubble? 0:16:09.280 --> 0:16:12.760 Actually, let me reframe the question. Do you think generative 0:16:12.760 --> 0:16:15.520 AI is a trillion dollar industry? Do you actually think 0:16:15.520 --> 0:16:17.360 it is the next hype of growth market? 0:16:17.640 --> 0:16:21.360 Let me answer that question slightly indirectly. 0:16:21.560 --> 0:16:22.080 Sounds good. 0:16:22.800 --> 0:16:26.800 I believe that generative AI has the capacity to add 0:16:27.280 --> 0:16:32.120 a trillion dollar or more over time if we use 0:16:32.160 --> 0:16:38.880 it correctly, because as an information technology it has great capabilities. 0:16:38.920 --> 0:16:41.880 We live in an age in which useful information is 0:16:41.920 --> 0:16:44.960 scarce all sorts of chunk you don't want is on 0:16:45.000 --> 0:16:47.920 the internet. But when you actually need to solve a problem, 0:16:48.240 --> 0:16:51.000 get better at what you're doing, get more background information, 0:16:51.120 --> 0:16:55.000 those things are very difficult to find, and generitive AI 0:16:55.080 --> 0:16:59.600 could be a tool for providing that sort of information 0:16:59.680 --> 0:17:02.360 to all all sorts of decision makers and workers blue 0:17:02.400 --> 0:17:06.159 collar workers, office workers and so on. But that's not 0:17:06.240 --> 0:17:09.120 the direction we're going in which case, I don't think 0:17:09.119 --> 0:17:13.280 it's going to add trillions of dollars of true value. 0:17:14.040 --> 0:17:17.359 But that also doesn't mean that generative AAI companies are 0:17:17.400 --> 0:17:19.560 going to go bust, right, because they're going to be 0:17:19.560 --> 0:17:22.840 able to monetize this in other ways. So if generative 0:17:22.840 --> 0:17:26.399 AI enables you to take over the search market from Google, 0:17:27.480 --> 0:17:30.760 that's a huge amount of money. It may take over 0:17:30.800 --> 0:17:34.439 the search market from Google without providing much better service 0:17:34.480 --> 0:17:37.879 to consumers, but it might still be hugely profitable. If 0:17:37.960 --> 0:17:42.480 GENITTIVAI companies convince businesses to invest in generative AI, that's 0:17:42.480 --> 0:17:44.080 going to be very profitable for them, but not so 0:17:44.160 --> 0:17:46.199 good for the businesses that misimplement it. Right. 0:17:46.800 --> 0:17:49.840 So the thing is, and I understand why you're making 0:17:49.880 --> 0:17:54.600 these assumptions, but what if it doesn't get cheaper because 0:17:54.680 --> 0:17:56.639 right now, the thing I've been on about with generative 0:17:56.640 --> 0:18:00.159 AI is, on top of not being super useful, it's 0:18:00.240 --> 0:18:04.280 so unprofitable and every report seems to be suggesting it 0:18:04.320 --> 0:18:07.840 isn't making people money. What if it stays where it 0:18:07.880 --> 0:18:10.119 is because in the last eighteen months, GPT four row 0:18:10.240 --> 0:18:13.960 is not significantly different. What if they've stalled. What if 0:18:13.960 --> 0:18:14.800 this is all we've got. 0:18:15.760 --> 0:18:19.400 Yeah, my guess is that it will get somewhat cheaper 0:18:19.720 --> 0:18:24.840 because right now it's very costly to even answer queries, 0:18:26.400 --> 0:18:29.439 and with more GPU capacity it will get somewhat cheaper. 0:18:29.800 --> 0:18:33.000 With better designs, it will get somewhat cheaper. But I 0:18:33.040 --> 0:18:36.000 do not believe that there is a scaling law in there. 0:18:36.640 --> 0:18:40.240 So many people in the industry believe in this mysterious 0:18:40.240 --> 0:18:43.159 scaling law, which is that you double the GPU capacity 0:18:43.240 --> 0:18:47.320 or comput capacity, you double the data, and you get 0:18:47.600 --> 0:18:48.560 twice the performance. 0:18:48.720 --> 0:18:51.240 Just an aberration of Moore's law by people who don't 0:18:51.280 --> 0:18:52.280 necessarily understand it. 0:18:53.160 --> 0:18:54.920 But first of all, what does it mean to say 0:18:55.000 --> 0:18:58.040 double the data? We're going to throw more Reddit at it. 0:18:59.480 --> 0:19:03.119 So even if there were such a scaling law, you 0:19:03.119 --> 0:19:05.960 would require high quality data, which we're not producing. Is 0:19:06.000 --> 0:19:07.600 run out. We're not paying for it. 0:19:07.720 --> 0:19:18.120 Yeah, there's something just very nihilistic about the whole thing 0:19:18.560 --> 0:19:21.520 as it stands. There's not really much of a social output. 0:19:21.520 --> 0:19:25.960 It's helping automate away jobs prolominantly held by contractors, which 0:19:26.000 --> 0:19:29.040 is already a problem onto itself. But also it doesn't 0:19:29.080 --> 0:19:31.560 seem to be making the money. I don't think I've 0:19:31.680 --> 0:19:35.760 ever seen anything like this in tech, and I'm just wondering. Indeed, 0:19:35.760 --> 0:19:38.480 maybe this is the question what happens if this is 0:19:38.520 --> 0:19:42.480 not it? But also TICK doesn't have a next step. 0:19:42.720 --> 0:19:44.600 Do you think one of these big companies could die? 0:19:44.720 --> 0:19:46.960 Do you think that there is actually an existential risk 0:19:47.119 --> 0:19:49.560 if generative AI and all this force apart. 0:19:50.480 --> 0:19:53.639 No, I don't think so. I think none of these companies, 0:19:54.320 --> 0:19:58.119 you know, are just committed to generative AI. They have 0:19:58.240 --> 0:20:01.920 other businesses that are making money, and even Nvidia can 0:20:01.960 --> 0:20:04.119 make still a lot of money with the GPUs. 0:20:04.280 --> 0:20:06.840 Let me rephrase it then, So right now, all of 0:20:06.880 --> 0:20:09.280 these tech companies, they do very well in their multiples 0:20:09.320 --> 0:20:12.160 in the markets because they have a relatively low cost 0:20:12.200 --> 0:20:15.040 of goods, like their actual costs are pretty great, but 0:20:15.080 --> 0:20:18.840 they're predicated on this ongoing growth. They must always grow. 0:20:18.880 --> 0:20:22.080 But what happens if they don't have a new growth 0:20:22.080 --> 0:20:26.679 thing because they haven't for a while, and what if 0:20:27.040 --> 0:20:29.320 they turn on generative AI? Like this feels like this 0:20:29.359 --> 0:20:31.400 could be an economic panic onto itself. 0:20:32.320 --> 0:20:35.200 Yeah, it could be. It could be some drops in valuation. 0:20:35.960 --> 0:20:39.720 The general pattern we have seen with many other products 0:20:39.760 --> 0:20:42.240 and technologies is that it looks a little bit like 0:20:42.280 --> 0:20:46.600 an S curve. Right, you're an acceleration and then you plateau, 0:20:47.040 --> 0:20:50.960 and that's when new products are invented, new investors move 0:20:51.000 --> 0:20:54.400 on to other things. And that hasn't happened with tech. 0:20:54.840 --> 0:20:59.879 You know, Microsoft is living its fourth life or whatever 0:21:00.119 --> 0:21:04.359 since a mestos, partly because they have acquired new businesses, 0:21:05.400 --> 0:21:10.320 some competitors, some competing technologies, and sometimes some tech companies 0:21:10.320 --> 0:21:13.399 have invested in the wrong things. I mean, cryptocurrency was 0:21:13.520 --> 0:21:16.920 more crazy than AI. There. I really didn't see the 0:21:17.040 --> 0:21:17.639 use case. 0:21:17.840 --> 0:21:19.800 The question I keep and have asked a lot of people, 0:21:19.840 --> 0:21:23.159 this is just what happens if there's nothing though, because 0:21:23.160 --> 0:21:27.480 growth is slowing. There is a pattern of slowing growth 0:21:27.480 --> 0:21:31.520 within these companies and there isn't a new thing that 0:21:31.520 --> 0:21:34.400 they can pick up and acquire. I don't know whether 0:21:34.480 --> 0:21:37.440 tech has ever had this happen is the problem. 0:21:37.680 --> 0:21:41.120 Yeah, it's a good point, but it's even deeper than that. 0:21:41.280 --> 0:21:45.480 Growth has slowed in the industrialized world, and it's not 0:21:45.520 --> 0:21:49.040 a new phenomenon. Since one of these paradoxes, which needs 0:21:49.080 --> 0:21:51.960 to be repeated more and more, the tech age has 0:21:52.000 --> 0:21:55.080 also coincided with a slow down of aggregate growth and 0:21:55.160 --> 0:21:59.600 every indicator of aggregate growth. So we are growing much 0:21:59.680 --> 0:22:03.200 less today than we did in the seventies or sixties. 0:22:04.280 --> 0:22:09.359 Productivity is growing less, and I think this is also 0:22:09.440 --> 0:22:12.320 related to the fact that we're not getting enough out 0:22:12.320 --> 0:22:14.440 of the new technologies and the new ideas and the 0:22:14.480 --> 0:22:18.399 new scientific discoveries that we are making. And part of 0:22:18.440 --> 0:22:21.960 the reason why there is so much hunger for AI 0:22:22.200 --> 0:22:25.240 hype is that many people, including policymakers by the way, 0:22:25.400 --> 0:22:29.520 are wishfully thinking, oh, well, this could be a solution 0:22:29.640 --> 0:22:32.720 to our productivity slow down. So perhaps in the next 0:22:32.760 --> 0:22:36.240 decade we can have a much faster productivity growth thanks 0:22:36.240 --> 0:22:37.480 to generative AI or thanks to. 0:22:37.480 --> 0:22:39.720 AI right, new jobs and such. 0:22:40.359 --> 0:22:42.680 Yeah, new jobs, sudden discoveries. 0:22:43.200 --> 0:22:46.240 So almost history is kind of slowing down. I've not 0:22:46.320 --> 0:22:48.560 really heard anyone really discuss it in these terms, but 0:22:48.600 --> 0:22:51.160 it's interesting. So you've seen that this is the growth 0:22:51.200 --> 0:22:54.320 of all costs is everywhere, and growth is slowing. But 0:22:54.440 --> 0:22:56.760 it sounds like growth isn't just about a money thing though. 0:22:56.960 --> 0:22:58.840 No, no, growth is not just about money thing, and 0:22:58.880 --> 0:23:01.640 I think if you do look at other indicators, we're 0:23:01.680 --> 0:23:05.680 doing worse. One of the regularities of the twentieth century 0:23:05.840 --> 0:23:10.280 across the world is that health and life expectancy have 0:23:10.359 --> 0:23:16.680 improved every today. People in Sub Saharan Africa have twice 0:23:16.720 --> 0:23:19.720 the life expectancy at birth as people who lived in 0:23:19.760 --> 0:23:23.800 London or Manchester in the eighteen hundreds, and Americans have 0:23:23.920 --> 0:23:28.280 had tremendous improvements in life expectancy and health until the 0:23:28.359 --> 0:23:31.480 last decade when it's slow. Then it started getting reversed. 0:23:31.800 --> 0:23:34.960 So on many indicators were actually doing even worse than 0:23:35.160 --> 0:23:37.240 GDP suggests. 0:23:37.680 --> 0:23:40.920 So what's contributing to it? Is it a welfare issue, 0:23:41.000 --> 0:23:42.480 is a societal one? 0:23:42.560 --> 0:23:46.040 Is it? Well? I don't think there is a clear answer. 0:23:46.119 --> 0:23:48.400 Some people think it's because of you know, the life 0:23:48.400 --> 0:23:51.919 expectancy part is because of early deaths due to alcoholism, 0:23:52.200 --> 0:23:55.720 opioids and drugs, But there is a more general deterioration 0:23:55.880 --> 0:23:59.000