You're listening to Asia Centric from Bloomberg Intelligence, the podcast that pulls back the curtain on global business so you can invest better across the Asia Pacific rim. I'm John Lee in Hong Kong, and I'm.
Kaite Dmitrieva with Bloomberg News, also in Hong Kong. We're looking at AI this week and specifically how it might not be this big growth driver that markets and some analysts seem to think it might be. On the high end, McKenzie sees up to a twenty six trillion dollar boost to the global economy. The top economist at Goldman Sachs sees it adding nine percent to US productivity.
Yes, but not everyone is this optimistic. According to Sequoia, AI companies need to earn at least six hundred billion in revenue to justify current AI infrastructure spend. Seems like we're going nowhere near this figure, and our guest is also critical.
I'm excited about our guest this week, John, because he's one of the big thinkers on AI, both in terms of the social and economic impacts. So to walk us through this brave new world is darron a Smoglu, Institute Professor at MIT and author of books, including Why Nations Fail. He's joining us from Cambridge, Massachusetts. Welcome Deren, Hi Katia, Hi John.
It's great pleasure to be with you to talk about AI.
We're glad to have you. Darren. You found in a recent paper that AI will impact something like less than five percent of all tasks. And you pulled back and you looked at the effects on the US economy, finding that US productivity would rise by only about zero point five percent and GDP growth by zero point nine percent over the next decade. So it's a lot of numbers there, but it's overall just quite a bit lower than estimates from many of your colleagues and economists tell us about that.
Why do you think the impact will be less than what people think?
Yeah, I mean, I think the bottom line is exactly. It's a bit, quite a bit lower than other numbers that are being floated around, and I think it reflects a deeper issue, which is that we are very much at the beginning of the AI technological journey, wherever that might be, wherever that might go, and there aren't many
applications that can be transformative. Yet if you look at AI what it can do right now, it can provide a little bit of better information to a few decision makers, and it can perform a few tasks, but it does not have the capability to do much in any task.
In any occupation that involves a central element of interacting with the real world, such as those in construction or manufacturing, or those that involve moving things around the real world, it cannot at the moment have a big input into things that are social in nature, for example, psychiatry, entertainment, things that involve many individuals coming together and using their
judgment or team interactions. So once you exclude all of these occupations, there are a bunch of white colored things that people do in their offices that require better information, that can benefit from better processing of language or data, and those are the things that AI can at the moment have a moderate, non trivial, but a moderate effect in helping us do better. And that's at the root of both my numbers and the fact that we don't have the killer apps for AI.
Yet, Darren. If you listen to a lot of people on AI, they say it's going to be transformative for most industries, but you seem to take a different view, like what percentage of jobs do you think AI will have an impact?
Well, I think that's a very difficult question to answer, John, because it depends on what horizon we're talking about, and what types of investments and what type of direction of AI we choose for the future. So when people talk of absolutely transformative effects of AI, I think they are mixing two things. One is the capabilities of AI as we have it today, and the second is where AI might go, for example, something close to superintelligence. It's a
hope or nightmare of some people. And of course, if AI is so smart that it can do many many things, and it can be integrated with robots, it can start driving cars and airplanes, that is a very different story than the kind of AI.
We have today.
The other difficulty in answering that question is that there is sometimes a presumption that somehow there's a single direction of AI, that we are just moving towards that direction, and the only choice we have is whether it's going to be China or the US would gets there and how quickly we're going to get there. The truth couldn't be farther from that. There are so many different things we can use AI, like technologies. At the end of the day, what we're calling AI today is machine learning
large capacity for effectively processing data. We can use that for automation, we can use that for science, we can use that for creating better information for a variety of different occupations. So depending on which direction we take and how quickly corporate investors are convinced to jump on the AI bandwagon, its effects are going to be different than it's effects that are going to be more or less pervasive.
I mean, there's certainly investors who have jumped on the AI bandwagon. I guess the question is have they done it too soon? Yeah?
Absolutely, I think, Katya, that's the key question. The reason why I wrote the paper that you mentioned, Katya, is because I think the AI hype is very counterproductive for two reasons. First, if you look at the narrative in the United States in the second half of twenty twenty three and in the first half of twenty twenty four, you will see that managers in many companies, not just publicly traded companies, but in many companies are under tremendous
pressure to invest in AI ads. New journalists, their colleagues, management consultants are continuously on their next saying what have you done on AI, are you falling back behind your competitors, And the outcome is inevitably people throwing money into AI without knowing what they're going to use it for, nor having the technology be ready for doing useful things for
most companies. The second is that the AI hype, fueled by a few tech companies and a few tech journalists is actually cementing the current direction of AI and the current structure of the industry. Open AI wants trillions of dollars of investment. Why because they believe they have the right business model and they want to be the leader. They don't want anybody else to catch up, and the best way of doing that is actually exciting investors to
want to invest more in AI. But first of all, it may well be that global society is going to be better served by a different technology, more open source, or a different approach to AI, or even if we're going to adopt the same approach, we may want more competition. We may want smaller startups to sort of be the ones that grow. So those are all things that investors attention and focus are going to influence majorly. But even more importantly from the point of view of my research,
is the direction of AI. I think that the current emphasis on automation and using AI for manipulating users on social media or better ads and better sort of ways of capturing consumers is not the most socially active one. So if we need to change the direction of AI in a more socially productive direction, we may again need to go a little bit slower and more contemplatively about what is it that we want to do. What are the places where AI can have the biggest social impacts?
And again the hype doesn't help that.
Can we pull back? I'm curious, you know, we're talking about what investors expect from AI and how much companies are investing in it. What are we talking about? Is it just automation? What is AI in business today? And it seems like a simple question, Oh.
My amazing question, because it's very difficult to get a straight answer to that. If you talk to some business leaders, especially in private, they'll say, of course, we're going to automate everything. That's what we want to do, and that's what AI is for Others, Sometimes the same people more publicly say no, no, we don't want to automate. We
want to use AI for other things. If you look at Microsoft, which is the big partner to open Ai, they brand everything Core Pilot, and I think it's a genuine effort for them to try to find a way to use AI to help managers or to help decision makers. But on the other hand, if you look at what open AI's leaders say, they say, we're going to automate
to stuff. So you know, it's a real confusion, and I think that confusion reflects exactly the same forces that we've been talking about Katya, which is that a we don't know where the future lies. There are many directions, the technology is malleable, we can make different choices, and right now we are very much at the beginning, so
we can dream. I personally think, for example, that all of this talk of superintelligence and you know, AI doing everything that humans do better than humans within twenty years, etc. It's all misleading, but there's no way we can know whether it is or not, because we're talking about twenty years in a very fast changing field, and people can dream.
Asia Centric is produced by Bloomberg Intelligence. We're more than fi five hundred experienced analysts and strategists work around the clock to bring you timely, world class research. Our coverage spans two hundred market indices, currencies, commodities, and industries, as well as over two thousand equities and credits. If you like what you hear, don't forget to subscribe and chairm Yeah, Darren, there's a lot of people like an ai to building railroads.
There's a view that, like nineteenth century America, if you build the railroads, the trains will come. Sounds like you don't buy into that.
Well, look, that's a very interesting debate. John perfect analogy because I actually personally think railways were critical for US development, and they were pretty pretty important for British development. But you know, some of the most respectable economic historians who work on this topic, for example, Robert Fogel, who want the Nobel Prize for his work on railways, made his
reputation by arguing exactly the opposite. He said, Look, we had other ways of shipping goods between places and transporting other stuff and moving people around, such as canals, and if you look at railways, that improve things. But the marginal improvement wasn't all that big. So he argued that railways weren't such a big deal. Now, I think that railways really did change inputs into manufacturing and change the
nature of technology. So I'm not agreeing with Fogel, but I'm just pointing out that even on something so important that happened one hundred and fifty years ago, there is disagreement. So of course there's going to be some disagreement on AI as to whether it's going to be completely transformative as a hub that changes everything, or it's going to enable us to do a few things that we were doing a little better look at search. Of course, I
think AI can help search. You said just the ideal kind of thing, which is, you know, it's an information processing type of technology. So when I typed something into my browser, I can now get help from AI. But is that going to be transformative? If I get a little bit better suggestions about where to go for on vacation, or whether I get pointed to the right journal article, is that really going to change the nature of truism? Is that really going to change the nature of science?
So those are the questions that I think we need to grapple with.
And what would you call transformative change? Like you sell you some examples where you know, having this this app maybe suggest a different kind of cover letter or you know, suggests a different The different introduction to a paper is one thing, but transformative changes something else transformative.
So let me give you two examples from my vantage point of transformative technology that I think most people will be familiar with, and then I'll tell you what my direction for AI that will be transformative. I think one transformative technology was Henry Ford's car factories, because they completely reorganized manufacturing and made it possible to first of all, use electric power in a much more efficient way, at the same time automating work and introducing very new range
of new skills and new tasks. And that really spearheaded transformative growth in the auto industry and also in other manufacturing industries that copied it. Another one is the Internet. Sure there was a hypen, there was a boom for the Internet, but if you look at the Internet, it was a rather new way of bringing people together. It was at the same time an amazing new technology for
sharing information. But it also enabled companies to often completely new services, create new platforms, So that I think, in my mind is transformative. So in the same way, I think AI as an information technology can be transformative if it does two things. One is just like the Internet. It helps us create new platforms to bring people together, to exchange formation, to exchange labor, to find better ways of actualizing their potential. Second, it actually helps us deal
with fundamental shortages in the economy. So today we have a great shortage of skilled crafts people, electricians, plumbers, better manufacturing workers, more skilled teachers. I think these are all things that can be helped with AI. Because why Because AI can act as a tool that provides information to people that's relevant, real time, context specific. So you can be a better electrician with AI helping you. You can
be a better teacher with AI helping you. And the critical thing here is that if you really want to realize it's not automation. You're not replacing the electrician or the teacher, but you're trying to enable them to do better in their task and to enable them to perform new, more sophisticated tasks. I think those are the paths for AI.
And you see something in common between these two directions, I pointed out they're both about amplifying human capabilities, not sidelining humans, but augmenting humans, creating better tasks, new things for humans to do.
Do you use AI?
I used it. Yeah, I used early on chat Gipt, and I would be lying if I said the first time I used chat Gipt if I wasn't pretty impressed by the way that it converses with you. But soon I discovered that I could do most things that chat Gipt three point five or chat Gypt four was capable of in other ways that I felt more comfortable, like, for example, if I went to sources, I could check these things myself faster rather than you know, converse with AI,
get some suggestions, and then follow the leads. So at the end of the day, I am not currently directly using AI, but I'm of course aware that in some other platforms that I use, AI is in the background. So I use Google for search, and I use an Apple phone and there is some AI in the background.
They're an outside of chatjpt. There seems to be a dath of any killer AI apps. Have you seen anything getting excited over and how much patience do you think investors will have finding the new killer app for AI?
No, I have not seen anything that I would say it's a great app, And I don't actually think that chat Gipt itself isn't that great. Again, I think the capability is there. If it was used differently, the architecture of AI and the vast amount of processing power and data that it has could be more useful. But I
don't find it so impressive. If you know, chat gipt can write a shakespeareance on it, or can sideline teachers and pretend that you know, all the students need to do is actually go to chat gipt and ask questions. Those are not to me killer apps. So I think that's the next big task, next big challenge for the industry. Let's find something that's both useful for businesses and actually socially beneficial. That would be such a good challenge for the industry.
Yeah, the social benefit of it's sort of the idea that I want AI to do my dishes and fold my laundry and not take my job away.
That would be great too.
Yes, I wonder if we could move outside of the US. I know you focused on the US and your paper, but you do have a global outlook, and I wonder if the impact of AI might be felt differently in different countries, depending on where they're starting from and the penetration of technology. What are your thoughts?
Absolutely, I think there are a couple of issues there we have to watch out for. The First one is that what happens in the US and other leading AI powers is going to influence what happens in the rest of the world. The rest of the world is not ready for AI. They're politicians, their thought leaders are not focusing on AI, so by and large, AI is going to be something that's done to them rather than something
in which they have agency. Still, it can turn out to be good for them because if it goes in a direction that really makes you know, a worker's more productive, then it will spread to say India and Indonesia, and it's going to make workers productive there. But I think if it goes in an automation direction, things are reversed. Many developing countries critically depend on their human resources semi skilled, cheap, active,
flexible labor. If AI starts replacing things that this word labor does, it's going to have, you know, disruptive effects. If AI changes the global division of labor is going to have disruptive effects. And moreover, many new technologies create a pattern of winner takes all, meaning that some countries that are first may leave the rest behind. And that's another concern and then the final wildcard, well, which is we're already seeing it realize in the developing world, is
that AI is also a phenomenal technology for surveillance. You know, a lot of the AI energy in China is going to surveillance activities. That's been very well documented. And also many of the surveillance related technologies, ranging from facial recognition, things that can sweep the Internet, et cetera, are just are not staying just in China. They're being exported to dozens of countries around the world, and most of them authoritarian,
and they're using these technologies with great enthusiasm. And it's not just authoritarian countries. I mean, you know, we can debate where US should be on the spectrum, but US invests a lot in AI related to national security apparatus, and you know, privacy from the government has become less secure in the United States and in some other Western
countries as well. So there are going to be other politically first order implications of AI as well, and those might matter more for the developing world.
Okay, but it sounds like a reading between the lines, sounds like developed, rich countries would have a bigger benefit from AI than developing countries.
I think right now that would be my guess. But if I had one recommendation to developing country leaders, I would say, this is the right time to form a new consortium, you know, especially with leading countries such as India, Indonesia, Brazil, Turkey, Mexico playing a leading role which is about AI, and there needs to be a perspective from the developing world and a voice from developing world, because you know, we hear from businesses in the US, we hear from increasingly now,
which is a good thing, from some worker groups in the US or in the Western world, but we don't hear the perspective, the interests, the sort of the things that the developing emerging world once and is concerned about AI.
I wonder if if we see that happening yet, is that happening?
No? No, I mean, if you go to many developing countries, there is interest within the public and some businesses, but there's no policy. Policy makers are not focused on that. And unfortunately, you know, we live in a polarized world that's divided, not just within countries. It's not just within the US or within France that there are big divisions.
There are big divisions between countries as well, and so it's becoming increasingly difficult for different country leaders to work together with each other.
Yeah, there's you mentioned the US in China earlier. It seems like there's this new I mean, not a new analogy, but you know, the space race, right except take it to AI between China and the US. Is there a sense that you know, given the minimal economic impact that you've computed so far, I mean, is there a risk that they're spending on the wrong things.
Well, they're spending on the wrong things, and it's maybe actually fueling the wrong narry. You know, two arguments that are very common in the US are First, we cannot regulate AI because if we do, China is going to take the leadership. And two, we have to be tougher on China otherwise they will become the AI leader, and we want to be the AI leader. And whoever becomes
the AI leader controls the world. Now, what you see in both of these narratives is not just you know, hyping up AI, but also deepening the tensions between US and China, which is the last thing we want, Which is the last thing we want in a world without AI. And if we're gonna have AI, we want it even less because why because AI is a global technology, So if you're gonna regulate it, you need US and China to work together.
So if you're an investor right now, like we've been talking about some pretty big themes, if you're an investor and you are putting money into a company on a bet on an AI bet, are there risks to that? Should they be doing that?
Investor? And of course there are risks, and it's a very difficult thing to do because even imagine everything I said here is right, and it's even more severe than I am saying. But as long as investors keep on investing in AI and Vidia, stocks are going to do well for the next year. So I think for many of the AI companies are AI related companies. What matters is really the market's focused for the next two, three,
four years, And that's really really difficult to understand. As John Maynard Kinges said, the stock market is like a beauty contest. It's not what the true value of the stocks are that's important, but what other people think the value of the stocks, the values of the different stocks are, So that's the same thing. So if other people think
in Nvidia is very valuable. If other people think that in Vidia chips are going to be a high demand, that's all that matters, not whether and Nvidia chips are going to revolutionize the world or not.
And what would it take for you to change your mind and get more bullish on AI?
Great question. I think if I saw AI really capably perform more tasks than I am envisaging, if I see you know, AI write articles or do news programs as good as you guys, or if I see AI teach students in a way that can form the same social bond, and then those teachers the students do mentally well, they perform well in tests and have reasonable sort of growth retention. You know that is going to be a real shock to me.
So, Ron, are you working on any new projects?
Yes, I am working on a new book which I think also AI and all of this data based economy raised questions about what is our relationship to technology as humans? How does it change what we want to value and what we want to do as humans, which is both an economic and philosophical question, so I'm trying to explore which sounds very interesting.
It's been an intriguing discussion on generative AI and whether they can meet very high expectations. It's been a great conversation. Thanks Darren for coming on the show.
Well, thank you for having me on your program, Katia and John and it's been a true pleasure.
I'm John Lee in Hong Kong, and I'm Katy Dmitrieva, also in Hong Kong.
This podcast was produced by Clara Chen and you've been listening to the Asia Centric podcast