Are We Building AI for Progress or Power? — ft. Daron Acemoglu

0:00Support for the show comes from Anthropik, the team behind CLAWord. When you're analyzing market trends or trying to understand what's really driving economic shifts, you need more than surface level takes. Meet Claude, the AI thinking partner that works through complexity with you. Whether you're dissecting earnings reports or exploring the ripple effect of policy changes, Claude helps you dig deeper into the analysis that matters. Try Claude for free at clade.ai slash propgmarkets and see why the world's best problem solvers choose Claude as their thinking partner. at blinds.com it's not just about window treatments it's about you your style your space your way whether you diy or want the pros to handle it all you'll have the confidence of knowing it's done right from free expert design help to our 100 satisfaction guarantee everything we do is made to fit your life and your windows because at blinds.com the only thing we treat better than windows is you.

0:56Visit blinds.com now for up to 50 % off with minimum purchase, plus a professional measure at no cost. Rules and restrictions apply. Not too long ago, everyone in digital media had the same idea. Get really big, really fast with help from Google and Facebook. Turns out there was a problem with this plan. There was no loyalty. These audiences were not really the media company's audiences. They were truly rented from Google or from Facebook. That's Brian Morrissey, the super sharp media observer and media operator. And this week on Channels with Peter Kafka, I talked to Morrissey about how to run a media business without getting big.

1:35What happens to the industry next? That's this week on Channels, wherever you listen to your favorite podcasts. Today is number 43. That's the percentage decrease in peanut allergies over the last decade. At a true story, I'm pretty sure I had a nut allergy when I was a kid. My parents thought I was trying to avoid church. Listen to me. Markets are bigger than us. What you have here is a structural change in the world distribution. Cash is trash. Stocks look pretty attractive. Something's going to break. Forget about it. Oh my God, that's pretty bad, right? I guess you're coming a mile away in all transparency.

2:15Brings up an interesting question, Scott. Do you have any allergies uh no i i literally am i as a matter of fact i can't i'm so sick of uh waiters are i think legally mandated in the uk i don't know if they are in the u.s to ask you does anyone have any allergies is that right yeah they have to ask you do you have any allergies i'm like bad service i've had it with the service in london you are so spoiled in new york the service is so good it's true what are some of your horror stories i think horror i mean you know i don't it takes some more in three minutes to get me my second makers and ginger those are my horror stories my my worst days are better than most people's best days but it's not i go to this place called since chiltern burned down i got to go to this place called kensington roof gardens which has way too many fucking dudes and it's way too crowded beautiful place though i i was there over the summer it's unbelievable yeah it's great for the 11 days a year you can actually go into the roof garden, but otherwise we're all crammed inside wishing the children was still open.

3:15And, and there's, I'm not exaggerating. There's 700 people lining up for the bar, 695 of the men. And you'd think, and there's like two bartenders sitting there, like slicing limes very methodically and elegantly. And I'm like, Jesus Christ, dude, just start like literally spraying beers at us and we'll open our mouths. That's the thing about, um, that's freaked me out about the members club you're a member of chez margot because now that we're paying it i'm not a member of that you're not i keep seeing you there yeah i get invited by by members well anyways the thing about it is it freaked me out i'm like it's so young here and then i realized it's not the reason i think it's so young is you know what it has most of these clubs have guys in their 40s and women in their 20s and 30s and what chez margot has is it has a bunch of dudes in their 20s which makes it feel like high school i i gotta agree with you surely surely surely that's a you're not very happy about i'm fine with that i like it they all come up to me i love your content on man hey what's that like and i literally you were you student body president or the mascot or something everybody knows you and it totally bums me out because i like to get a little fucked up and talk to everyone at the bar throw an umlit cigarette in my mouth put on sunglasses and just be a cliche of me and then when people come up and say they know you, I feel as if I have to act semi-respectable.

4:36I'm like, oh, hey, hi, yeah. Oh, yeah, no, I don't know what's... What do you think is going to be the next interest rate cut? God. God. They're all so fucked. This is the trouble. You shouldn't have been a professor. You clearly... You should have been like an actor or like a pop star or something. That's what you really want. A hundred percent. I'm just not that talented, but I agree with you. No, I think you are. You're just... You're talented in the thing. you've got a good brain you've got a good economic brain and really what you want to be doing is like you've got the bezos thing you want to be you want to be like justin bieber or something uh you had me till the justin bieber part i would have gone for ryan reynolds or george cluny he seems smart and very politically oriented but the thing the the thing i hate about your friends that's a good way to start a sentence is they're so they're so annoyingly earnest professor calloway it's so nice to meet you i know ed he's such a good person isn't he fuck off anyways they're so anyway it's but it's it's great to meet your friends out i've got to learn about who these friends are i haven't been your princeton buddies what is it like a reading club or a drinking club Like you guys have Pim's Cups or something and play cricket?

5:55Eating, yeah. Is that what it is? Eating club, that's right. So instead of fraternities, they have eating clubs? They actually have both. Oh, really? Yeah, that is what it's called. It's called an eating club. But you were in an eating club. And you would get into anyone because of that faux British accent, which, by the way, everyone is figuring out is total bullshit. They're onto me. You're from Nashville. All right, and enough of this. Get to the headlines. One of our douchiest. Okay, here is our conversation with Daron Asimoglu, Nobel Prize-winning economist, New York Times bestselling author, and professor of economics at MIT.

6:36Professor Asimoglu, very good to have you on the show. Thanks, Ed. Great being with you and Scott. So we want to start with AI for obvious reasons. That's all we can really talk about at the moment. We've been seeing a lot of bullish sentiment in AI. We've also been seeing a lot of circular investments happening. A lot of people saying that it's a bubble. Last year, you were asked to answer on a scale of one to 10, how much of an impact AI is going to have on the world. And your answer at the time was negative six. So we want to unpack this, starting with that hot take. What are your views on AI at this point?

7:23And do you still have a negative six rating on AI right now? Let's break that into three pieces. One is, is AI a transformative technology with great capabilities? Yes. That's why it's not minus one or plus one. It's minus six. Second, how quickly will this technology reach fruition? And there, I think the answer is it depends on how quickly it's pushed. So I think for a positive development path, I think we need a more deliberative approach. we are rushing into AI in a way that I think makes applications using AI less likely to develop because we are just doing it too quickly. And we also don't have a roadmap of what it is that we really want from AI, while everybody recognizes that this is a technology that's going to have tremendous number of side effects, foreseen and unforeseen consequences.

8:25So all of these things, plus, very importantly for me, the focus on automation and AGI, while there are better things to do with AI, I think tip the scales towards negative. So when it's minus six, minus five, minus seven, take your pick, but I'm very worried about the direction of AI where it's much more concentrated, who uses, who controls information and what we do with it. Yeah, could you break down what that negative impact would actually look like? There's the aspect of the concentration of power, and I'm sure that could have many implications. We're also worried about it. There is the implication of automation and the idea that this would replace labor, it would replace people's jobs.

9:09Say more about how it goes from negative one to negative six in your view. What is that destructive impact that you're so worried about? in the production domain, if we use AI mostly for automation, I think not only would we be missing some of the really transformative uses of it, but we would create much smaller productivity gains than expected. And we would also create various social outcomes that are negative related to job loss for certain groups, lack of employment opportunities for certain groups, wage stagnation or declines like we've experienced in the 1990s. All of these are on the negative part related to the production process.

10:04But I'm also very worried about the fact that AI is first and foremost a communication technology. And as a communication technology, it changes political and social dynamics. And when it centralizes information in the hands of a few companies, it can have a variety of very negative effects on democracy, on dissent, on diversity, a variation in opinion, all sorts of things that we are really not prepared for. What would you say to someone who would say that you are being a Luddite? Or, I mean, we've seen transformative technologies in the past, whether it be oil or the electric grid or the internet.

10:57And it seems as though when these transformative technologies come along, there is a lot of concern about what it will do to our economy and how it will negatively impact our economy. But many people would say, well, eventually it works out and we shouldn't be so worried about technology. What do you say to those people? Well, there are really two theories that people could have about long-run effects of technologies in general. The first one is that in the long run, things will work out by themselves. Let it rip. And just the dynamics of things, for example, in the labor market or via democratic processes, sometimes semi-democratic processes, is that we'll just find the right way of dealing with things.

11:51The second one is that, no, it's a deliberative set of choices that we have to make in order to make sure that the long-run effects are better than the short-run ones. And I think if you look at several transformative technologies, indeed, they did have fairly negative short-run effects. The beginning of the British Industrial Revolution is associated, depending on how you measure it, 70, 80, 90 years of real wage declines or stagnation and huge increase in inequality. The transition out of agriculture similarly at first created quite a lot of social and economic hardship. But in both cases, later adaptation worked out much better.

12:36But I would say even though decisions were made without a roadmap, there were specific decisions about how to use technology, how to change the organization of production, and also political decisions that were quite important. And so it's not an automatic process. So I definitely am not an AI pessimist. And so I don't know what your definition of Luddite is, but I'm not an AI pessimist. But I do not believe that we're going to get the best out of AI or even the second best out of AI if we just say, oh, let's not worry about all the disruptions. Somehow things are going to work out. So it feels as if a lot of the concern or discussion around AI has moved towards who can secure reliable, large amounts of electrons or energy.

13:25one is this an attempt it sounds like do you think this is an attempt to paint a future where the demand is going to be unlimited and a bit of a head fake or do you see the same sort of power constraints that these guys are seeing and what do you see as kind of the downstream impact of that i do not believe that power gpu capacity is the main limiting factor for the kind of ai that I have in mind. Because if we're going to make AI really serve our needs, I think it needs to have much more human complementary domain-specific expertise. It has to be able to be an aid to electricians, to accountants, to journalists, to academics.

14:17And for that, high-quality domain-specific information, data, is going to be the real scarce resource. Whereas right now, the energy demand is mostly from foundation models that are very impressive in some ways, but also very, very expensive to run and have not reliably reached that context-specific domain-relevant expertise. So I think we have to find a way of getting the best out of foundation models, but combine them with domain-specific models. What I've also seen is that a lot of the different LLMs are hitting sort of a technical parity by most kind of, I don't know, hardcore metrics. Do you think there's a scenario?

15:01We had Robert Armstrong from the empty on several times, and he said that there's certain technologies where no one or small set of companies is able to ring fence stakeholder and therefore shareholder value, The airlines, PCs, vaccines were huge innovations, but you didn't have a small number of companies garnering trillions of dollars in market cap. Do you think this might qualify as one of those industries where even if it ends up having a huge impact on society, that we're overestimating the ability of a small number of companies to capture a ton of shareholder value? I would be very worried about an industry that is so concentrated.

15:41On the other hand, it's a cutthroat industry. And we don't understand what sort of industrial organization of AI is going to emerge. So it is almost certain that whoever is doing the foundation models advances is not going to be the same one, same company that also does all of the applications. So you're going to form an AI stack. So once you form that AI stack, where are the risks going to reside and which part, which layer of the stack is going to get most of the returns? I think that remains to be seen. It's going to depend on where the real bottleneck is in terms of doing useful things and whether the foundation models are close substitutes for each other when it comes to serving as the first layer of that stack.

16:35I think those are really interesting questions. And it is made more interesting by the fact that industries that look very competitive at some point later on can be very non-competitive because early competition is about being the one that controls things later on. That's why it's so vicious because everybody thinks that they're going to get the prize and the prize is dominate the industry. So I don't know whether the competition that you're seeing right now is going to repeat itself in 10 years time, or whether we're going to go to a winner-takes-all sort of structure at the foundation layer.

17:14Yeah, in your book, Power and Progress, you basically take us through history and the history of technological progress. and you make the point that technology has not necessarily been as beneficial as we think of it because of how it has been distributed throughout societies, which seems extremely prevalent to what we're likely about to see with AI. Take us through that history. What is your reasoning for that? How does that play out? Yeah, I think there are essentially several reasons why the full potential of a suite of technologies may not be realized or may not be realized quickly. One is monopoly.

18:00If one or a few companies dominate everything and they use that in order to extract all the rents but also slow down innovation, that's one recipe for many good things not happening. So today, I think the digital world or communication world would be very different if AT &T had remained like the sole monopoly. So the breakup there probably opened the field for more shakeup. The second is, you know, whether that technology is working with labor or is sort of just replacing labor and worse, like during many parts of human history is becoming a tool for repressing labor. Those don't work out great.

18:51I mean, slavery was not a very efficient system. It wasn't just bad for the coerced people, but it wasn't actually generating economic dynamism. At the time of the Civil War, the U.S. South was falling further and further behind, while the number of patents, innovations, industrial production, and all sorts of other things were advancing rapidly in other parts of the United States. So I think all of these things we have to sort of take into account, is AI going to be a monitoring technology where workers become more and more powerless because there's so much data being collected about them? That's another concern that we don't often talk about.

19:32So there are many issues here that are intersecting, and we see parallels for each one of them in history. None of those are perfect parallels. We've never been confronted with a technology that's so widespread in its potential applications. But we've had other technologies that are quite transformative as well. So how do we harness AI in a good way? I mean, what does regulation look like in your view, such that it is a net benefit to society versus a negative six? First of all, I think we need to be thinking about what it is that we want from AI. Of course, not everybody's going to agree on that, but at least that conversation needs to be had in a more open, constructive way.

20:20And second, we also need to be clear about what we mean by regulation. I think what most people mean by regulation is a reactive kind of regulation. Something happens, AI companies do something, and then we are worried about certain additional harms, and then we regulate that. I think instead, what we want is something more proactive. Let's think about where it is that AI can do most good and ask ourselves whether it is going in that direction and what are the impediments for it not to go in that direction and see whether we can do certain things to facilitate that. So if we did not do those facilitatory things, we would not have the internet because the government support there was quite important.

21:04We would not have renewable technologies that are now, at least in certain applications, cost competitive with fossil fuels. So it's not like we have a sort of a law that says the market, or in particular, a few companies that are steering technology, are necessarily going to choose the right paradigm. I think the market is excellent in doing certain things, but market participants are also locked in a particular type of business model often. They are going after a particular kind of prize. And it is possible to step back and say, well, is there another prize that we should be focusing on? And that's what I'm arguing that human complementary AI, where we try to augment human capabilities, expand human capabilities, could have real benefit.

22:00And that's not the direction in which we're going.

22:05We'll be right back after the break. And if you're enjoying the show so far, be sure to give Prof G Markets a follow wherever you get your podcasts.

22:37It's a collaborator that actually understands your entire workflow and thinks with you, not for you. Whether you're strategizing your next business move or diving deep into economic analysis, Claude extends your thinking to tackle the problems that matter. For finance professionals exploring earnings reports or economic policy ripple effects, Claude works through complexity rather than rushing to conclusions. It helps spot connections across multiple sources, challenge assumptions, and develop insights that go beyond the obvious. Try Cloud for free at cloud.ai slash PropGMarkets and see why the world's best problem solvers choose Cloud as their thinking partner.

23:32all on one AI platform. No more juggling multiple systems. No more worrying about growing too fast. Just the full power of Workday helping small to mid-sized businesses like yours run more smoothly. And Workday Go activates quickly. You can be up and running in 30 to 60 business days. So simplify your business. Go for growth. Go with Workday Go. Visit workday.com slash go to learn more.

23:59Scott, we're hitting the road, bringing Pivot Live to the people. Seven cities, Toronto, Boston, New York, D.C., Chicago, San Francisco, and L.A., of course. You went to Oasis, you went to Beyoncé, you saw the remake of Wizard of Oz and The Sphere. All those suck compared to the Pivot Tour. This is the biggest tour. Same people that are organizing our tour that organized Taylor Swift's tour, they are much more excited about our tour. All right, that's enough, Grandpa. It's going to be so good. And we're bringing our brand of whatever we do to the people. And we're excited to meet our fans. We love our fans.

24:38For tickets, head to pivottour.com. See you there.

24:49we're back with prof g markets do you think the u.s is going to be able to maintain i mean other than deep seek it's just very difficult to think of another ai player of almost any importance globally outside of the u.s do you think that the u.s is going to be able to maintain that type of lead in the AI ecosystem? China has an engineering advantage. They have a huge number of engineers. They're generally well-selected, meaning that more talented, quantitatively sort of skilled people enter into engineering because it's a very prestigious thing, and they have exams that are relatively unbiased, and engineers are highly regarded.

25:30So when it comes to pure engineering things, I think China could have an advantage. On the other hand, the top-down system is hugely inefficient. There are so many places where inefficiencies build up. People are afraid of taking initiative. There is no decentralized sort of process. There, the U.S. has an advantage. How that will shake out at the end would really depend. DeepSeek is an engineering marvel. They didn't come up with any of the new methods. The new methods that DeepSeek was using, some of them were invented by Google and OpenAI. Many of them were invented 20 years before by machine learning scientists.

26:09So they just took them, but they combined them quite well. How many more times can they do that? That's going to be one of the questions. And then, of course, it's not just US and China. Can other countries catch up? Europe is behind, clearly. But I don't think there is a law that Europe has to be behind. There are many talented AI scientists in Europe that just happened to be all working in Silicon Valley. I think this is a good segue into your 2012 book, Why Nations Fail. We're discussing how America could fall behind in the AI race. And it's interesting, you're kind of highlighting that on the one hand, there are some benefits to this top-down communist structure where you can set the agenda and the tone for the nation, the tone being, we love engineers and we love people who build AI.

26:58On the other hand, it can be a problem when you don't have the competitive forces of capitalism at work. And it seems as though this is going to be sort of the defining question of our time, which one works. 100%. I think that's very well put. Which brings me to the question, why do nations fail? I mean, you've done Nobel Prize winning research on this, the role of institutions. Why do nations fail? It's mostly institutions. There are other factors, but many of these other factors, like just civil wars, are institutional as well. And the role of institutions, both formal rules, but also informal arrangements and norms, they become much more important when we're dealing with sectors that are forward-looking, innovative, require small players to scale up.

27:58all of those are things that the U.S. was doing quite well. You know, the United States had an ecosystem of startups that, you know, were extremely confident. You know, people, when they opened, I mean, I knew many of them. I know many of them. Very few people think, well, if I launch a startup and if I'm successful, will I be shut down by courts? Will I be crushed by my competitors? Will I be able to get any contracts when my competitors are favored by the government? Nobody thought about that. Nobody thought about that, partly because I think people were on the optimistic side, but largely because U.S.

28:40institutions had a pretty good track record of not doing that. I think we're no longer sure. There are favored companies and not favored companies. Courts are much less impartial. scaling up may become much harder when there is more uncertainty. So I think there are a bunch of issues where the institutional advantage that the U.S. economy had is more of a question mark today. And the problem is that when you mess up certain things, you pay the price right away. If you mess up institutions, especially as they pertain to innovations, you don't pay the price because the impact's not going to be felt for another five, 10 years.

29:21So if there are some fundamental negative effects from Trump's attack on, you know, independent judiciary, the costs of that will be realized not in 2027 or 2028, but probably in the 2030s. What is like the counterfactual to an institution-led society? Like when we talk about it's the societies that have had strong institutions. Those are the ones that have worked out. And that's what your research has explained to us. What is the alternative? What is a society that is not led by institutions actually look like? What enters the void? How does that lead to a less prosperous path? Well, I mean, I would think that every society has forms of institutions.

30:14But let's change your question to state-led versus not state-led. So Soviet Union was state-led, but too much state, not enough market, not enough decentralization, and that was horrible. But, you know, Somalia, where clans ruled for several decades after the collapse of the state, is the outer opposite. There are no state institutions that were functional. There's no third-party enforcement. is anarchic ways in which even small problems would escalate. Neither of these two are great. Probably the Somalia is worse than the Soviet Union for economic activity, although probably the Soviet Union makes up for it by killing people more effectively.