Immigration in the Age of AI Could Look Very Different

Today I'm speaking with Garett Jones, professor of economics at George Mason University and author of three major works that he calls the Singapore trilogy: Hive Mind, 10% Less Democracy, and The Culture Transplant. As a Singaporean, life here can sometimes feel like the movie adaptation of Garett's books, and as in many other cases the books are a lot more fun than the movie. He previously worked in the US Senate and is now the chief economist at Bluechip, a stablecoin rating agency. His pointed research on intelligence, immigration, and democracy might test the relationship between how we think the world works and our favorite political positions. I think we should take that test seriously, because in true Singaporean fashion it is a difficult test and one that many will fail. I want to hear how his ideas might have changed in the wake of AI and ever-shifting geopolitical dynamics.

Welcome to Second Orders, Garett. It's good to have you.

Great to be here. Thanks for inviting me.

Deep roots

Plenty of ink has already been spilled, and saliva sprayed, on your ideas on migration. Most of it has been critiques of your reasoning and evidence. I want to do something a bit different this time and spend more time unpacking the forward implications of what happens if the theory is right. Can you first give a brief outline of deep-roots theory and how it informs your views on migration more broadly?

In the early 2000s, a variety of papers were published by academic economists that looked at the long-run effects of migration on institutional quality and national productivity. Economists love natural experiments. We do not get to run real experiments very often in macroeconomics, the way people do in the medical sciences, but we do get to watch nature's experiments. One set of those experiments over the last five centuries was created by the great waves of migration after 1500. You can check and see: when a lot of new people moved to a place, how did things change?

Country of origin matters a lot. Places that received a lot of immigrants from Western Europe tended to be pretty successful. That is also true of places that received a lot of immigrants from East Asia, especially China — the great Chinese diaspora as part of a natural experiment. Places that received a lot of immigrants from sub-Saharan Africa tended not to do as well. These are broad generalizations, but they are empirically strong enough, robust enough, that they led to what is now known as deep-roots theory: the idea that people bring traits with them.

We do not know exactly how these traits are transmitted. There is a debate about how much of this is purely cultural, how much of it might be genetic, how much it might be some complicated fusion of the two, and that is beyond current tools for testing, clearly. But the idea that what is old is new again, or that the more things change the more they stay the same, is definitely true when it comes to migration over the course of many centuries. Deep-roots theory is basically a set of glib generalizations spurred by the research of people like Louis Putterman and David Weil, and William Easterly, who if I recall correctly is now at NYU.

The ideal migrant

If we just examine one dimension of a culture — and maybe the literature already covers this — I am interested in the idea of the ideal migrant, because one of the applications of your theory is that there must be, at the very least, an archetype of an ideal migrant. But even at a conceptual level, is it feasible to conceive of a coherent ideal migrant if you consider the multiplicity of problems a country has at any one time, and the accompanying multiplicity of possible human-capital solutions?

A deliberately crude example: if a country has an excessively low savings rate, getting migrants from a frugal culture would probably improve that country. The converse would be true. If you have a high savings rate, you probably want people who come from a more spendthrift culture. The difficulty is that a country might need people from a more frugal culture on one dimension, and on another aspect of the social fabric, or of the way they trade or conduct business domestically, need people a bit more willing to spend money. On that dimension there would be a net detriment. Must there first be a weighing — these are all the problems we have; some are more important than others? That is complex enough as a static example, and it is really dynamic. People change, countries' problems change, and the traits they bring to the table as solutions change over time. So how do we conceive of the ideal migrant?

Let me push back on your initial example. When it comes to savings, I think just more is better. Just frugality is better. That is partly because I and most macroeconomists focus a lot on the long run, and in the long run there really is no paradox of thrift — no world in which savings make you poor because people do not consume enough. In the long run, which is where we should keep our focus ninety-plus percent of the time, higher rates of frugality are going to be associated with higher rates of patience, more foresight, a better ability to predict the future. These are all skills that you want more of. No country ruined itself by saving too much. On that particular trait I'll push back. I just wanted to stand up for team macroeconomics and team Frank Ramsey, the great economist who taught this idea about a century ago.

But the general point is very sound. There are a lot of different things you want, and there might be some trade-offs between them. If we think of immigration policy as a way of a nation hiring new citizens, and we think of the government's immigration department as basically a huge HR department trying to decide whom to hire — it is hard to hire. It is hard to decide whom to hire. Anyone who hires knows it is a total crapshoot whether the person is going to turn out the way you think they will.

The fortunate thing is that all the good traits are positively correlated. Psychologists refer to this as a g factor, or a first principal component. People who are smarter tend to be more patient. They tend to be more cosmopolitan. They tend to be more liberal in their attitudes, including being more generally pro-market, at least pro-trade. A lot of these good things correlate together. At least at a weak positive level, sometimes a strong positive level, if you just selected on IQ, or if you just selected on thrift, at the individual level or the national level, you would do pretty well, because a lot of other good things would come along with it.

I do think we should be Hayekian enough to say we do not know exactly what we need, and there is a lot of unknowability in any social process. But one thing we do know is that smarter people with more pro-market attitudes on average tend to make the world better than people who are the opposite. I don't think the needs change much over time. Part of the reason we can tell that is that if you look at the global rankings across countries for the last century or so, things have not changed that much. The big changes have been when communist or dirigiste economies became more laissez-faire. That has been the big change: a bunch of people waking up and saying, oh yes, all this central planning was a disaster, and then institutionally deciding, with mostly the same people there, that they were going to become more laissez-faire, more Friedmanite in their economics. That has been the big rank-order change across time. It is not that high-trust societies turned out to be too fragile in a world of advanced internet technology that made it easy to prey upon people's trustworthiness. Weird tricks like that have not happened. This strange fact of persistence is the fact to explain and the fact to build upon when thinking about immigration policy.

Papa John's institutions

Is it just the cultures that persist, or both the culture and, as you mentioned, the environment and what kinds of traits the environment selects for?

The way most economists would put it who have not fully bought into deep-roots theory is that institutions are persistent. The rules of the game, government quality, legal-system quality: these things tend to persist a lot from generation to generation. Where I take this is that I build upon deep-roots theory and the associated term I call cultural-transplant theory and say: let's figure out where these institutions come from. A lot of that comes from the ingredient story. Who shows up? Who do you have in your culture? When you change the people, you usually do change the policies, and to a substantial degree.

The most direct answer is that in general, as long as the people are not changing too much, there is going to be a lot of institutional persistence, with these unusual once-or-twice-a-century big shocks — the wonderful effects of the fall of the Berlin Wall, where a bunch of people who bought into the wrong model made a good-faith effort to try to build onto a better model. So there are some non-deep-roots, non-culture-transplant elements of institutional quality, and fortunately in the last fifty years people have been awakening to realize their institutions are worse and improving them.

It could go the other way. We worry a lot about whether the egalitarianism of the West could end up planting the seeds of its own destruction. Joseph Henrich's The WEIRDest People in the World is a story in which critics of Western decadence say maybe the institutions of the West end up planting the seeds of their own destruction; their egalitarianism is now going to undermine itself. That could be a case where ingredients that worked well for a couple of centuries could end up becoming dysfunctional. Some people argue that the values of East Asia and its Singaporean diaspora could end up having some robustness that Western Europe does not have — that Western European egalitarianism and individualism could end up being the wrong ingredients for the twenty-first and twenty-second centuries.

There is a slight shift there. If on the one hand we are talking about the persistence of institutions, it is one thing to say that an institution is persistent because of qualities or traits that the people who populate those institutions, expand them, and operate them have, and another thing to say that those cultural traits are themselves persistent. Is the institution persistent by way of the persistence of the cultural traits that compose the institutions?

I think it is overwhelmingly the former. Yes. There is a pizza chain in the United States known as Papa John's, and their slogan is better ingredients make better pizza. I refer sometimes to the Papa John's theory of institutions. Better ingredients make better pizza: better people make better policies, and worse people make worse policies. That is why getting the raw ingredients right for your society is of first-order importance. Things like intelligence and frugality and pro-market attitudes are those ingredients that will tend to build better institutions. I am open to the idea that there could be some pathological feedback mechanisms there, but if so they are rare, and they are rare enough that we should keep our eye on the main ball: traits like intelligence, frugality, and pro-market attitudes.

If better ingredients make better pizzas, is there any world in which better pizzas in turn end up making better ingredients? We can make the causal link from people and cultural traits to institutional quality. The primary focus of the chain of causation is from people to institutions. But on what time horizon, and under what circumstances, would you say that the feedback loop between institutions and people in that reverse direction is stronger?

I do not know if the second, reverse causation is stronger, but it is at least important. The big example would be things like education and health. If you have got good institutions, then even people who come from cultures where education and health systems are really weak reap all the benefits. When Western Europe gets a lot of migrants from sub-Saharan Africa, these are folks who come from societies that have, on average, very poor educational systems and poor health systems. What do they and their descendants get? Better education and better health. They end up being healthier, more productive members of society. If you are not fighting a bunch of intestinal diseases all the time, you are going to be a more productive worker. If you have an educational system where the teachers show up, where the teachers have not sold their books in order to make money on the side, you are going to get a better education. That shows up to a very important degree. Good institutions are good for everyone. They do a good job lifting folks who would have wound up at the bottom of their own societies up closer toward the middle. That is a big benefit. The modern rich countries are good at creating widespread human flourishing, at devoting resources to those who are difficult cases and helping to improve their situation. The old glib answer was something as simple as eyeglasses. There might be some people who just could not read easily if it were not for eyeglasses, and eyeglasses are basically free in rich countries.

Evolution, diversity, and hybrid vigor

I want to pick up institutions later and return to the ideal migrant, or the perfect migrant. One of my bad habits is relating evolutionary theory to everything.

I am fine with that. If there is one thing we have learned in the last century and a half, we have figured out evolutionary theory, and we have figured out there was something beyond Newtonian physics. Those are the two big additions.

How sympathetic are you to the parallel between evolutionary theory and deep roots? The specific parallel: if naturally occurring mutations are critical for having a large enough sample size for the better mutation, the mutation more adaptive to the environment, to be found and then to achieve victory over the others — if that is a critical piece of the puzzle — then wouldn't narrowly or preemptively saying we are getting only migrants of a certain type from a certain place disqualify us from getting access to that specific cultural trait which a migrant might have? Ten years ago this was not a problem we were facing, so we allocated more resources to getting migrants from other countries, and now the environment changes, or some black-swan external event changes the environment, and we are ill-prepared in terms of human capital. Does this parallel make sense, and does it present any problems for deep-roots theory?

It is a reasonable question. The first response would again be coming back to this idea that there is a first principal component, that the good things are all correlated with IQ, or something very closely correlated with IQ. Here is the way I'll put something I learned from psychologists. If you can find a positive trait that is negatively correlated with intelligence, and you can publish that in a journal, you will instantly become a famous psychologist. Every decade or so I will run into a famous psychologist and ask: is this still true? If I find some great cognitive trait, or maybe even a non-cognitive trait, that is negatively correlated with IQ, would that make me a famous psychologist overnight? The answer is always yes. So the idea of there being a systematic relationship, even a weak negative correlation, between the thing you are selecting on and the thing you really want is a theoretical possibility, but it seems rare enough that there is no reason to sweat it.

The more important thing is that there are plenty of ways to have a diversified immigrant population that do not require saying we should be negatively selecting on intelligence. That is why being open to immigrants from a lot of different countries is a good idea. That is one way to get a valuable form of diversity.

My colleague Jonathan Schulz, at George Mason, in my department, an excellent researcher, published a paper that kind of reinforced things other people had found: that the great migration America experienced in the late 1800s, where a lot of European migrants from various parts of Europe moved to the United States, improved innovation, and it looked like it was diversity per se, not just more bodies showing up. Schulz uses simple indexes of immigrant country of origin, and those indexes from well over a century ago predict innovation, patenting, whatever R&D-type stuff you are looking for, well over a century later.

If I have to say what is one thing worth trying to break up when you can, it is basically raw ethnic homogeneity, if your goal is innovation. Little cartels, little clubs, need a little competition. One thing economists are fans of is competition, and they worry about monopolies. Ethnic monopolies can end up becoming ideological strongholds, monocultures, and bringing in enough competition in the world of ideas or culture is probably a net win for society as a whole. Frictions in the short run can turn into a lot of dramatic novels that people worry about — oh, the Italians and the Irish, how are they ever going to get along? I am sure there are versions of this in Singapore. Those conflicts are very likely worth it. I'll go beyond "probably," because cross-fertilization is really important.

That is one element of the evolutionary analogy I can bring this back to: hybrid vigor is a pretty strong result in evolutionary theory. Hybrids can help you avoid deleterious recessive mutations popping up in the population if you just bring in more diversity. That form of increased diversity is probably a win for innovation, and since I am a big fan of trying to boost the level of innovation in the world, I put a premium on that even at the cost of some short-run conflict.

Diasporas as instruments

I want to move on to how, in practice, states respond to deep-roots theory. I have heard you say that further work has to be done on this, that we might not have enough case studies. Has anything changed? If we think about states that are deliberate migrant exporters — the Philippines, Bangladesh, Nepal — and even involuntary net migrant exporters, countries people are fleeing, conflict-stricken places with refugees: how are these states likely to react? If a state knows it is high-value per deep-roots theory, do we see them start using their diaspora as instruments of influence? Conversely, how do migrant exporters who are bleeding talent respond if they knew that other high-value states were trying to get their people?

Let me break it into a few pieces. First: are countries with high-value migrants likely to be able to pull off using them as instruments of influence when they move to the new culture? I think the answer is basically generally not. Countries are always going to try to spy. They are always going to try to manipulate things. They try. But we should keep in mind that most of this stuff just does not work. You end up paying a lot of money for intelligence services and they tell you things you can read in the New York Times. Every once in a while I get to meet someone who has a fairly high security clearance, and they are often very excited to read a new book that was just published. If someone who has access to the greatest secrets of the most powerful nation in the world still thinks, oh, I should go read this New York Times bestseller, that tells you there just are not that many secrets worth knowing. That is stage one.

Stage two is to what extent a country is going to be able to control its migrants a generation or two down the road. There is always a possibility of that, but that will be a weak effect. Every generation we have heard about migrants in the US, at least in my lifetime. You hear about whether Russian scientists who come to the US will take information back to the old USSR. You worry now about whether Chinese-diaspora scientists who come to the US will end up bringing information back to China. I guess you are supposed to worry about that. But there is so much information flowing around that I think that is very much a second-order concern. People assimilate. Links weaken.

If you are really worried about this, then you should give these folks citizenship more quickly rather than putting them on green cards or other temporary visas where there is always a pull back to the home country. A green card or citizenship solves a lot of problems, because people actually like to have US citizenship or Singaporean citizenship or Japanese citizenship, and once you have that as an anchor the pull of the home country is much, much weaker. There are complicated situations with families, but family ties probably weaken quite quickly once you have a new home. That sort of indirect-influence story is high-status to talk about and of low importance in practice.

A point system, and a truly great replacement

The classic idea of brain drain is most commonly understood as an interplay between the home country and its people who are becoming migrants. What is different here is not so much "my home country is not so good and there are better countries out there, so I will uproot myself." If deep-roots theory becomes a consensus, and everyone knows there is this spoken or unspoken global hierarchy that determines which types of migrants, and by extension which countries, produce high-quality migrants as per deep-roots theory — how would this knowledge, this new social fact, influence the immigration market? Does it mean that if I come from a high-value deep-roots nation I should be able to be more selective about the countries I go to, because now maybe the countries are competing for my residency, as opposed to the current paradigm where a lot of people are competing to get into the US?

There already is a competition for high-skilled migrants by conventional measures. What deep roots or culture-transplant theory adds is that country of origin is its own superpower. It has its own predictive value above and beyond whether an individual migrant has a high IQ or good skills or an engineering degree from a fancy school. The two are going to be correlated a lot.

Here is the story I often suggest. My hunch — I have never done this quantitatively — is that the optimal immigration policy is going to look something like a point system where two-thirds of the weight goes on individual traits and one-third of the weight goes on country-of-origin traits. That is probably the optimal level of integrating deep-roots theory into immigration policy, and that would especially be true for rich countries. The poorer the country gets, the more they should place the weight on the deep-roots measures and less on individual measures.

Rich countries have already sort of figured this out tacitly. The big lesson is that the poor countries should be taking advantage of these ideas. If Madagascar decided that deep-roots theory was really important as a path to getting richer, they would say: we would take just about anybody from China, as long as they agree to come here, have some kids here, plant down their own deep roots, and stay for a few generations. The reason I picked China is because it has a billion people and a lot of them are not too rich right now.

It is the poor countries that could benefit most from taking deep-roots theory seriously. They should be subsidizing, or giving preferred tax status, or whatever you want to call it, to try to bring in folks from countries with higher deep-root scores than them, in order to improve their institutions — not even ten or twenty years from now, forty, sixty years from now. That is where I think the real play is for immigration policy: for poor countries to decide we are going to actually, actively create a culture transplant, create a great migration, create a great replacement, a truly great replacement. This term gets used as a derisive term in Europe. But I think there really are opportunities for dramatic demographic change that improves national policies in ways that are very likely to work out well in the twenty-first and twenty-second centuries. I do not think people will do it, because I think bigotry, ethnic rivalry, and people's intransigence, people's unwillingness to try new things, are going to get in the way. But it is obviously possible to pull this off.

I see a cruel paradox of deep-roots theory, and why some people might feel it smells a bit reactionary: it seems dead on arrival for poorer states. If you are a poor state and you need someone from a deep-roots high-value country, there is very little you can do to incentivize that particular individual, and very little you can do at scale, because you are poor in the first place. Maybe you have some beaches, but besides that not much to offer. It leaves us with countries that are doing well getting people from good countries, and countries that are not doing well not being able to get those people, self-perpetuating extreme ends that never converge again.

My one push, mentioned in The Culture Transplant, is the excellent book China's Second Continent, written by a New York Times reporter who traveled, who had strong ties in China and in sub-Saharan Africa. There are apparently about a million migrants living in sub-Saharan Africa now. Some of it is elite stuff — we are building dams or electric programs — but a lot of it is just people building little small manufacturing plants, running little shops, running a restaurant, a nightclub in some medium-sized town.

It would be difficult to scale that up. How would you go from one million to ten million to fifty million? Many countries have a little bit of leeway with institutional change. A lot of countries in sub-Saharan Africa, for instance, have gotten their act together on monetary policy and have had very good low-inflation experiences over the last decades compared with the past. There is some room for institutional reform if you make it a focal point. If Madagascar said we can get ten million folks to come in here from China over the next twenty years, and we are going to credibly promise not to expropriate them if they get rich — they might be able to pull something off like that. I am more than half hoping that some philanthropist, perhaps some philanthropist made rich by the IPOs of the great AI firms, will decide they want to create a new project of subsidizing immigration from China to one or two countries in sub-Saharan Africa. I am very confident that that is likely to turn out well fifty years from now, and people would look back on that and say that was a world-transforming and a world-improving change.

You can fake a résumé

A seemingly trivial thing, but it has the potential to be quite influential on how immigration markets change. This might seem a highly contrived way of bringing in AI, but one of the primary contributions of agentic AI is that it can make communication between any two human entities much more efficient, even if neither, or only one, of the entities is actually present. It can communicate on a human's behalf. Leave the technical complications aside and say it is a widespread technology: every human has an AI-agent clone twin. Do we see a possibility that, in the immigration process, people negotiate directly with the customs or immigration departments of other countries for the particular terms of their migration? If you are from a high-value country, perhaps as part of that negotiation you are allowed a little more leverage on the terms of your relocation than someone from a lower-value deep-roots country. Does communication at scale converge with deep-roots theory to create a new kind of immigration market?

One question is whether there is any information beyond the basic three or four facts of deep-roots theory. When I have suggested what tools I would use for implementing my ideas, I actually have not recommended using the deep-roots measures themselves, which would be fine, but I always say: maybe we use a three-value index of your country's national average test scores, your national average savings rate, and your national average level of corruption — the inverse of that. Could a fancy AI come up with something vastly better than that? I really do not think so. These three things have a pretty high R-squared, a pretty high correlation, and they are correlated with each other and with all the good things that you could actually measure. Everything else beyond that is kind of noise. There is very little else to negotiate about.

There is a reason a lot of legal contracts are written with clear dividing lines. There is a reason bankruptcy is defined in a clear way: you miss a payment, you are bankrupt. These simple enforceable rules are welcome to both sides of the transaction ex ante, because you want systems that are easy to enforce ex post and that have clear dividing lines. Human beings love a drama. We love nuance. Our brains are wired for it. But when we say how do we want to live on a day-to-day basis, we want to just say: you broke the rule in a really clear way; we are not going to worry about the nuance here.

Partly it is because there just is not that much information, and partly — the most pro-AI critique of what I just said — the information-processing costs are too high to deal with all the detail and the nuance, and maybe AI shrinks that, makes it a lot easier to handle those complicated cases and to negotiate over them. I have strong doubts, more than just some doubts, about whether AI-created legal systems that can negotiate at high speed and pull in vast amounts of information will come up with legal-settlement questions more broadly that are very different from the ones we already have. If that is true more generally, I am just applying this as a particular case study.

Part of my story is that when you are bringing in a migrant, you are bringing in their descendants too. There is going to be so much diversity among the descendants. Children and grandchildren are very different from the parents in some ways. The apple does not fall that far from the tree on the basic levels of intelligence or hard work; these things are kind of heritable, genetically and perhaps culturally too. But there is so much diversity about all the little details, which is where you have pushed the discussion: do I want somebody who is more of an engineering person or more of a literary person? Do I want someone who is warm and cuddly or bureaucratic? Those are things you cannot select for beyond the first generation in any material sense. The deep-roots stuff is about picking the migrant now whose children and grandchildren are going to help make your country great.

Here is why I asked. Your migration ideas, I think, sit in the lineage of models of the world in which, when you go down the abstraction ladder and become more concrete, one layer of concreteness below can negate or undermine the assumptions of the previous layer. You would be saying it is not just migrants from anywhere; it is migrants from countries with these specific cultural values, or these g factors. On the other hand, one of the benefits of communication technologies, and of AI in particular, is the specificity with which human interactions and activities can be performed: our increasingly weaker reliance on general-purpose technologies and general-purpose rules and standards, and being able to have everything customized. One incarnation of that is social media and the algorithmic feed, in that everyone has their own bespoke set of information streams. If these two things meet, it could mean something very interesting for the development of new human activities and also markets. AI would enable assessments at an individual level. What your argument, or what deep-roots theory, shows is that if you get more specific and more detailed, more nuanced, you get better outcomes, because you are better able to triangulate the relative value of one migrant as opposed to another. Both are saying specificity is better and specificity is possible, which only encourages me to extrapolate to the extreme and say: what if all immigration applications could be personally given 100 percent scrutiny and attention by an AI? Wouldn't you then undermine deep-roots theory? Then there is no such thing. We do not have to use IQ as a proxy anymore. We do not have to use any of these cultural traits as a proxy. We could use them as frameworks, benchmarks against which the AI will assess an applicant, but for all descriptive, non-normative matters the AI can elicit all this information directly from an applicant and does not have to rely on proxies.

Remember, the AI is stuck with the same knowledge problem that we all have, which is that basically people are good at faking things. That is part of the reason why it would rationally be skeptical of the claims in a five-hundred-page packet. If you reframe the question as what would the AI in its dream wish to have as a complete set of information about a person — suppose there were a credible way for me to just press a key and tell the AI you can watch the last six months of all of my social-media feeds, of everything I have ever typed, of all of my emails, and let's even suppose, falsely, that I would not have faked everything in advance in order to maximize my application chances — in that sort of utopian case, what is still going to happen is that it is going to figure out there are just a few big factors that are highly predictive of success, and they are going to look like a g factor, and they are going to look like some cultural traits that are correlated with my ancestry. They are going to notice that people from Confucian cultures are kind of different from people from WEIRD cultures, from Western Europe. The first factor that is going to matter a lot is going to be something like IQ. The second factor is going to be some sort of one or two broad cultural vectors. Beyond that there is a bunch of noise. I think an AI would land on that.

Second, there is the bigger problem of the fact that people fake things. When they know they are being evaluated, they try to rig the system. One of the benefits of using the deep-roots factors or culture-transplant measures, the ones that have been widely used in the literature, is, as I have said before: you can fake your résumé, but you cannot fake your country's résumé. Part of what you are worried about in any kind of immigration policy is to what extent people are just padding things, faking things, making sure they get a disability waiver so they could take four hours on the test instead of the normal two. These moral-hazard problems are universal in economics, and of course they are very important in immigration policy, as they are in any high-stakes decision. The simplicity, the hard-to-fakeness of these deep-roots measures, is one reason why I think an AI that had broad access to data would still place a lot of weight on the measures that you personally cannot control.

IQ, AI, and the O-ring

I want to move on to the links between national IQ and economic performance. When thinking about this and how AI plays into it, I thought of two approaches. One is that AI changes the nature of the relationship between national IQ and economic performance. The second is that AI breaks the relationship altogether: a complete severance, decoupling IQ from a country's economic performance.

On the first way: you could say the nature of the relationship changes if you are talking about complementarity and substitution, something you have discussed. Take complementarity: AI automates all of the lower-skill jobs while leaving the higher, more g-loaded tasks for humans to do. Does that allow higher-IQ countries to pull ahead even further? And the substitution scenario, wherein expertise is still crucial in what you call O-ring sectors or O-ring jobs because they require high-quality labor? Is it substitution or complementarity, and do you think of it as changing the nature of the relationship between IQ and performance, or as a complete decoupling?

A core part of the Hive Mind story is that IQ matters mostly through externalities, through spillovers, and a big part of that, maybe half, is this institutional factor. Smarter people build better institutions, which help the whole system work better. One question is: will AI be allowed by humans to actually build better institutions? That is the one scenario I can imagine where AI does really weaken the relationship between national IQ and national productivity: a situation where a lot of poor countries decide we are just going to turn over our government, and a lot of our management decisions for society, over to AIs. We folks who are not the sharpest tools in the shed — no human is, but we know that our test scores are not that great — we are just going to outsource a lot of our governance to AIs. In that setting, any country smart enough to do that, that could do that credibly, would plausibly within the next decade start seeing dramatic increases in growth.

My hunch is that there will probably be a couple of experimental middle-income and poor countries that will be full AI-maxers, and part of that will be handing over a lot of governance, routine bureaucratic decisions, over to AIs. Countries that do that will start seeing economic miracles that are visible to all, too obvious for anybody to ignore. That would be an electronic way of doing what I was suggesting before, which is having the Chinese diaspora move into your country and start doing a bunch of great stuff. AI may offer a socially acceptable alternative: instead of having the Chinese diaspora move to your country, you have OpenAI move to your country.

It cannot just be, oh, they are going to take over this little job or that little job, or we are going to have thirty AI agents running an assembly line. It has to be: we are going to handle the legal system so that people know in advance there is going to be very little corruption now, and we know that the rules of the game are going to exist for a long time and they will be transparent enough for real humans to understand them. One could hope that the sort of ethnic conflicts that have happened between humans when people of different ethnicities move to a country will not happen between humans and AIs. We already know that there is a tension between humans and AIs, but I am open to the possibility that there is enough intellectual and cultural diversity in the world that some cultures will be like, yeah, those AIs, they are fine. Yeah, I am okay having AI be 20 percent of our national assembly, of our national legislature. I am okay having it be our constitutional court. Those are the kind of situations where you can get narrowing in incomes across countries.

The flip side, which you also suggested, where you could see a dramatic widening, is where that cannot happen. People in poor countries cannot make the credible commitment to having the AI be their form of government to a substantial degree. In those cases, having AIs be part of these O-ring processes — these fragile, delicate production processes where getting everything right is super awesome and getting one thing wrong destroys all the value — that is, for the audience, the example of the O-ring, the space shuttle Challenger, which exploded in flight because one rubber O-ring was fragile. You had the entire ship working well, but one little piece malfunctioning meant that everyone on the ship died tragically. Many modern production processes are like that. Building computer chips is that way. Every step has to be perfect, otherwise it is worthless; you throw it away.

A lot of modern value is created through O-ring processes. It is very plausible that AI-led, AI-organized, AI-managed O-ring processes will be vastly more productive than those with humans in the loop. If it is currently the rich countries, where test scores are high, that are best at managing this stuff, and if they start being able to manage those kind of delicate production processes much better, then that is where you get a scenario of twenty, thirty years of the rich countries pulling away from the poor even more. The poor countries are basically running on 1990s technology and the rich countries are operating on science-fiction tech.

Which is interesting, because then this could be an answer to how poorer countries could attract high-quality migrants: if poorer countries tend, for whatever reason, to be the ones more likely to embrace AI in governance, that could be the incentive for several segments of the migrant population. It is a good experiment at the very least, and I will move there to try to see what life could be like.

That is actually a good point, because you probably need some sharp people to manage that process, and you probably need a lot of them. We have seen over the last three years that demand for skilled programmers has actually risen, not fallen, because they are a strong complement to AI implementation, AI improvement. We already know that there is a huge consulting industry in basically bringing AI into firms. It would be world-changing if we had a similar move of bringing AI into governance. The culture is not there for that. Government people are totally behind the ball on AI as a rule, and the exceptions to that rule are people basically in my greater orbit. The Mercatus Center has had a bunch of folks who are great on AI. Dean Ball is a person who has helped bridge that gap in an important way between government and AI. There are 180-plus countries in the world. If five of the poor ones just get on board with AI-maxing and bringing in the workers who would complement that as a part of bringing AI into actual day-to-day governance, that is going to be world-changing in a way that would be hard for the world not to notice.

Which is funny, because it just made me think about how the "AI took my job" discourse could then be applied to these migrants, the pioneering migrants who went to a poor country to build the AI governance systems and then automated themselves out of their jobs. If intelligence and g factors are the ones most deterministic about the relative economic success of a country, then as the dictator of poor country A what I will do is incentivize all these smart cookies to come build the AI and then deport them, because that is essentially what I needed the smart cookies for.

Fortunately, "AI took my job" is already a strong enough meme that the dictator would need to find a way to credibly tell the immigrants, we will let you stick around. The question is what sort of contracts the dictator could write in order to get the migrants to come, knowing that there is this ex post incentive for deportation. It could be that you just have to pay them a massive amount of money to get them to come. Or it could be that there is some kind of contract enforced in Switzerland that pays out a massive amount if they are deported, where you are basically outsourcing that payment to a third party. The meme is already there, and if the meme is in the programmers' minds, very few of them will move without some credible promise of fair treatment ex post.

Even with credible guarantees, there is always this information asymmetry between new country and migrant. Usually the idea is that a migrant from Indonesia knows a lot about the United States, and that is partly or chiefly the reason he wants to move to the US, but the US does not know enough about the migrant. Deep-roots theory helps fill in that gap by evening that information asymmetry. What you are discussing here is divergent in a sense: the migrant does not know, in the fullness of time, what would be the outcome of their transition to a poorer country if it was on this particularly idiosyncratic basis — I come here to help you build AI, but you must guarantee me a place afterwards. The information asymmetry persists in a different way.

I am really glad to see that the economics-of-AI discourse has penetrated a fair amount of sophisticated discussion. The new theory of the firm, the economics of organization that economists built up in the seventies and eighties, and for which there have been a bundle of Nobel prizes, really is going to inform how AI gets integrated into normal corporate governance and then, perhaps over the next decade, into actual democratic or even autocratic governance.

Lagos is not San Francisco

I want to ask how you think about the effects AI has on the place premium: the idea that there are particular places that enable people to perform much better, at least economically, than they otherwise would if they were in some other place, or more directly if they were in their home countries. Could surrounding a worker in Lagos with several AIs that replicate exactly the intellectual environment of San Francisco enable that worker to make as much as he would have if he had physically moved to San Francisco? And if not, what does that say about the relative importance of cognition versus institutions?

This goes back to the Hive Mind story: it is the externalities of intelligence that really matter a lot, and the core externalities are what we normally call institutions. Smarter people tend to build better institutions. But beyond that, in the old literature there was always this blurry distinction between institutions and culture, and I think it is time to bring back that blurriness. Part of the value of being around smart people is partly that they build better government in this formal, juridical way, and also that they build a better cognitive culture.

I am a big fan of the late philosopher Jim Flynn, who documented what we now call the Flynn effect, the idea that intelligence has been rising over time. Flynn did not just document that. He also had a theory of intelligence. Over the course of the twentieth century, life in the rich countries became more like an IQ test. It became more about looking for general principles, looking for levels of abstraction, looking at the ability to draw insights from the general and apply them to a new case. The old saying about intelligence is that intelligence is what you need when you do not know what you need. Flynn argued that a good cognitive environment is part of what makes you smarter.

To take a place like Lagos and create the same cognitive environment that you would have in San Francisco would involve having AIs around you all the time, having them be, say, half the people you talk to. One could create that, and then the AIs would basically be creating a San Francisco-level hive mind, at least on the day-to-day thinking level, and that would obviously, I think, raise cognitive scores, test scores, IQ scores, general subjectively felt smartness in Lagos on a high level. Social media must be doing this to some small degree, if everybody is watching first-world reels or TikToks. That helps some.

Beyond that is: how do you get San Francisco-level governance? Part of the reason San Francisco-level governance is so high is because it draws on the backstop of the American legal system. You could produce a lot in Lagos if you had something that was approaching rich-country levels of government quality. People are always worried: how am I going to get something done? There are a lot of links in the chain to basically building a chip-fabrication facility. It is not just having the factory there. It is getting things delivered in a timely manner, not having rent-seekers predate upon you at every step of the way. That requires a long production chain. That means changing all of the institutions. To get one good thing done in Lagos, you need to get a lot of good things done in Lagos. It does become that chicken-and-egg problem.

You would imagine, as you are probably thinking, that this could happen for simple call-center-type work or laptop-job-type work. People should be able to outsource that, and to some extent some of that must happen. But if we are thinking about anything beyond those mundane laptop jobs where people could be digital nomads — when we are thinking about long production processes — it really means getting a lot of things right, and that means getting things that are at the level of what we normally call institutional change. That is what makes it hard to take on grand tasks in Lagos that are at frontier levels of productivity. It is the long chains, the long production chains, the long supply chains, where many things have to be done right in order to get anything done.

Basically one way to put it is that everything is an O-ring. Everything is an O-ring production process. For the audience, this is Michael Kremer's great idea — K-R-E-M-E-R — Nobel laureate now: the O-ring theory of economic development. A very simple model, worth googling if you have not looked it up. It is very useful for thinking about what frontier production is like. You have to get a lot of things right in order to get anything done.

Basically the idea that inputs have multiplicative effects on outputs.

Yes. Exactly. And truly multiplicative. If you have a 99 percent chance of getting a process done right, that is great. But if you have a 99 percent chance of getting one step done right and there are a hundred steps — wow, not very much is going to get done right, even if it is a 99 percent chance of success at each particular step. A lot of production processes, a lot of value creation, are these long hundred-step production chains where the difference between 99 percent and 99.9 percent is not 0.9 percent. It is like 50, 60, 70 percent ultimately. Kremer used the example of a good that can get sold at a store like Saks Fifth Avenue, a very famous American high-end brand, versus what gets sold at their remainder store, Off 5th. When something is a remainder, the smallest little flaw in a piece of high-end clothing means it gets sold for half of its price or one-third of its price. With computer chips it is even more extreme. Those are just a small illustration of the kind of value gaps we see when production processes are this rigorous.

Would it be fair to say that the synthesis between cognition and institutions, insofar as your views on place premiums are concerned, is that a lot of things must be done right in order to get anything done — that is the argument for institutions — but in order for things to be done right in the first place you need high levels of competence and ability — that is the argument for cognition — and this is how the two meet?

Exactly. I think that is exactly right. People use the place premium in a lot of different ways. It is used as a general term. My story, a lot of people's story, is that most of the place premium around the world is about institutional quality. Most economists stop at that point. Where I go is: well, most of the institutional differences are due to human-capital differences. There are other place premiums too. San Francisco is just a nice place to be. It is beautiful. The weather is nice in some ways almost all the year. I used to live across the bay at Berkeley for a year. Pretty views are part of the place premium. But when we look across the planet, most of the place premium in my view is human-capital externalities, human-capital spillovers: being around smart people.

After ASI

Say we have an AGI, ASI scenario, where individual human IQ becomes economically irrelevant to production, because on the one hand a 130-IQ person who can design a cancer-saving drug is able to do the exact same things that an 80-IQ person could do because of AI. Is that a world in which national IQ becomes an economically irrelevant metric or proxy for other states to use? And in general, will there ever be a point at which we collectively say intelligence is not so much a requirement, or not such a valid proxy for how good a migrant can be for a country — either because there are other things we do better than AI at that point, because we just are not nearly clever enough, or maybe there has been some global equilibrium reached? Economic success is not a goal for many states. Maybe this is an issue of time horizons, how far we are willing to go. Where do you sit on this?

It turns a lot on this question of what the ASI ends up looking like, this superintelligence. But if it works sort of the way we have been discussing here, the key question is whether the people in the society are able and smart enough to credibly delegate a lot of decisions to the ASI. The ASI is basically a god that can solve many problems very quickly and give solutions, and perhaps run an army of robots that can make a lot of things happen to solve a lot of human problems. Then the core question becomes: which humans are willing to delegate authority and a lot of decision-making power over to the ASI, and which humans are not? Let's assume that still matters. Let's assume that people are still in a position to credibly make this decision if they want to. In that case, what matters is: are people smart enough and patient enough to let the ASI run things, knowing that there will be bumps along the road and bad things along the way, or does their short-sightedness, their impatience, their anger at the machines keep them from delegating stuff to the AI?

A lot of that is going to be correlated with intelligence. Not perfectly, but at least to a high degree, I think. Patience is something that is correlated with intelligence. Foresightedness is. I think that will continue to be important. The level of intelligence of a country will continue to be important in that kind of world, because the smarter groups are more likely to come together, to pick the focal point, and say yes, we are going to delegate a lot of manufacturing processes off to the AI. Yes, we are going to delegate a lot of government decisions off to the AI. Other folks are going to be basically species-centric. They are going to be pro-human and they are going to celebrate human values and not letting these AIs run their world, and they are going to be the poorer for it.

The strange thing is that the people most likely to espouse those beliefs and actually enforce them probably tend to be the smarter people who, whether by way of education or just overthinking, just being trained in the bad habits of overthinking, overly complicate the adoption and deployment process of AI governance. Whereas you could see maybe a poorer country with much higher rates of impulsiveness being just like, hey, let's dive head first and let's do this whole AI thing. We do not have to have our committees and debates and Substack blogs debating the pros and cons. So it could be that we see the first wave being the poorer, not-so-good countries, and they pull ahead because there is just so much waffling around and quibbling on the side of the more developed, smarter nations. Is that a possibility?

We hit on this earlier. Poor countries have the most to gain from the right immigration policy. Poor countries have a lot to gain from turning over a lot of big decisions to AIs. The question is whether they are going to be able to come together and do this. I really hope that a few do, because it will very likely provide, within a five-to-ten-year window, very clear signs of success, something as obvious as the changes that happened in Central Europe after the fall of communism. Within a decade it was really clear. Between 1990 and 2000 it was clear: wow, Poland is a lot richer than it used to be. Wow, what happened? And it is kind of obvious. A similarly rapid, obvious set of changes would be clear if some poor countries could credibly hand over some authority to the AIs.

The credibility problem is hard. But the fact that poor countries have done a pretty good job with central-bank independence over the last decade, and not getting into these terrible traps of heavily politicized lending and whatnot — the fact that they have improved that is a sign that there is a good chance of this happening for AI-led governance as well. It is speculative, obviously, but the improvement of poor countries in monetary policy is a big example of delegating authority to a bunch of nerds and just letting them decide what is going to happen on really important financial decisions. The fact that many poor countries did that is a hopeful sign that they might do the same thing with AI governance.

I see the segue into 10% Less Democracy there. But the point I actually wanted to make was a different one. When you were discussing how important it is to be intelligent, or at least have the traits positively correlated with intelligence, in an ASI world, we were focusing on that part of the timeline where we were transitioning into an AI society or ASI society, in that we must make key decisions that will have lasting and enduring effects on how the nation is administered, which parts of the government to deploy AI in and how to do so, which stakeholders to work with. This is true during the transitionary period. But if we take a stable post-ASI period, do you see a possibility in which, because all these things have been settled — whatever the optimal amount of government that should be done by AI is already being done by AI, and likewise for the private sector — human IQ is obliterated as a relevant factor in terms of economic success and performance altogether, across countries? That there is much less inequality across countries?

That is a possibility if human beings within regional geographies can still keep up that commitment to delegating things to the AI. That is the key question. You are sort of assuming the answer when you say suppose we are in a post-ASI future and we solved all these problems. That is the big question: can we solve these problems?

The rich-country version of this, on a small, almost comically, tragically comical note, is watching what is happening with Waymos right now, robot-driven cars. The cliché is that in the United States, Republican states get Waymos and Democratic states get studies. The desire to protect Uber-driver jobs is strong enough that it beats out the desire to save lives from car crashes. That is a really big decision, and it is important for people's lives, and yet there are a whole bunch of places that are getting it wrong, and they may get it wrong for ten or twenty years, maybe forever. Rhode Island just passed a law — I do not know if it became enacted or if it is just being debated — to require that a third of all grocery checkout counters be run by humans. The pushback against handing decisions over to these robots is going to remain with us for a long time. If the difference between ASI governance and human governance is 10x or 100x, then even small carve-outs for humans could be very socially costly. That is what I would want to think about: can societies credibly commit to ASI governance on stuff?

We could imagine a world where a century ago people said, I do not trust computers to decide whether to clear my checks. I want a human being looking at my check and making sure that I actually wrote it out for 1,700 bucks or $2.50. I need humans to check this. This is super important. If we lived in that world, we would be living in a completely different financial world than we do now. You can imagine some states, some places and times, saying we need the equivalent of human-inspected check clearing. We need human-driven cars. We need human-handled governance. If this truly is an O-ring process, if there are a hundred steps that are being done by the AI almost perfectly, but we are still requiring one step to be done by a human, in an O-ring sense this can be incredibly socially costly. So even if the differences across societies shrink from the point of view of us as people today looking out and saying, well, gosh, you are all being governed by robot gods, so who cares whether the robot god is handling 70 percent of your life or 99 percent of your life — in an O-ring sense, that could be a 10x, a 20x, a 500x difference in quality of life. The AI carve-outs, whether they are intentional or sort of unintentionally chosen, could end up having massive consequences for the wealthy nations.

Under the desk

AI and democracy: do you have a working model of how democracies evolve, and where exactly in this evolution technology plays a role? Can you extrapolate this to account for how widespread adoption of AI would change democracies?

We already know that everybody in government is using AI all the time, because they are just people and everyone is using them all the time. My guess is that in the US, half the members are checking it all the time. Maybe two-thirds, three-fourths. What is this bill about that I am voting on? That kind of stuff. So they are already using AI a lot, but at the same time their jobs depend on getting reelected. We are running representative democracy, and in representative democracies the vibes of the people are really powerful, and those vibes are not going to change very much anytime soon.

We are now at the point where we know that lawyers are using AI all the time to submit their briefs. That is probably going to raise the standard of the median brief, even though some people are still falling for hallucinations. The quality of certain kinds of technical work will increase. Levels of raw human understanding could go up or down, basically based on the substitution-versus-complement angle, and I do not have a good answer to that right now. So the quality of inputs gets higher, but at the crucial decision point people may get lazier, because they are outsourcing a lot of thinking to the AI. On net, under the current regime, I think that still improves governance, because people will use AI as a tool to help them improve their jobs, and from the point of view of elected politicians, they are doing their job better if they are getting reelected at a higher rate, and improving some other ex post benefits like keeping up relations with lobbyists who could give them jobs later. AI improves governance quality on net right now, even with the lazy substitution that occurs along the way.

The question is: at what point will we start formalizing the role of AI rather than just having everybody secretly holding it underneath the desk during the meeting to check it out, or integrating it into their laptop by using the software I will still not mention so they can improve their live chats? Formalizing the role will be the big step. Deciding that one or two or three of your Supreme Court justices are going to be AIs: that is going to be the big step. That is real outsourcing. My guess, based on past experience, is that people will not do it that nakedly. It is a little too obvious, a little too on the nose, to just say we have two Supreme Court justices who are now AIs. So they will have to find a backdoor way to integrate AI into democratic governance. I do not know what those institutions will be right now.

One piece of good news for AI and democracy is that so far, at least, there is one experimental study showing that AI seems to reduce belief in conspiracy theories. If everybody is checking their AI for stuff to see what they believe, it probably will reduce belief in some of the most extreme conspiracy theories, which is good news for the inputs for voter quality. That is one case where, without assuming broad institutional change, we probably get better governance over the next ten years as more and more people are checking their AIs all the time to say, should I really believe the screenshot that Elon Musk just retweeted? That probably improves the quality of discourse right there.

Short time horizons

What about short time horizons? That seems to be an important factor you consider in the wide array of political failures, and how insulation is the remedy for the problems brought by short-term horizons. Does AI change things when it comes to a technocrat's or a state functionary's time horizons, and does it change things when it comes to a voter's, the electorate's time horizon?

My immediate response is no, because that is an institutional-change question. The way you get the benefits of long terms is by delegating big decisions to some independent org, and that is what my colleague Jim Buchanan would probably call a constitutional-level decision. It is a big political change to decide we are going to go from two-year terms to four-year terms. Voters' demand for responsiveness from their political systems is pretty strong, and their willingness to delegate to distant bureaucracies is constrained, limited, idiosyncratic. AIs could in principle do better. In principle AIs are good at reminding you: here are the long-term consequences of this short-term impulse that you have. But how do you get modern representative democracies writ large, the big-tent version of representative democracies, to embrace the long-term thinking that an AI, and that nerds like me, are in favor of? That is an institution-quality decision. That probably takes a decade of persuasion — unless one sees that one or two countries have a huge success with AI-driven, with AI governance, and then other countries just feel like they have to keep up with it just to compete. That force of competition, that force of an example, can get people to change, not quite endogenously. You can get endogenous change from the success of a competitor. If a smart country like Denmark makes a decision to just outsource a couple of bureaucracies to AI and gets really big, obvious, beneficial results clearly and quickly, then other people start feeling like they have to imitate it. Otherwise I do not think we get that long-run thinking just from AI per se.

At the risk of beating a dead horse, I return to the idea of AI as the chief communications technology, or the last incarnation of communication technologies we might need — or maybe not last; telepathy would be cool. Do you see a future in which we have a lot more ballot boxes for a lot more issues, and that changes time horizons indirectly by allowing, in America, the primaries, or in other parliamentary democracies the general elections, to be focused on the merits and ideas of the individual as opposed to being issue-driven elections, because there is no need to campaign on issues, because issues can be voted on individually? You do not have to vote in proxies of an individual or a party. Every voter can just submit their personal preference for a particular issue. If that is the case, that might change the way campaigns are run, and how campaigns are perceived.

This is a story where AI reduces transaction costs for democratic input, and so people start to disintermediate. They say, okay, I do not need you to make the decision. We will just have a bunch of votes on things, and we can run it so quickly and cheaply that we will do it all the time.

This is something that has been possible for decades. People were wondering if the internet by itself would lead to big increases in plebiscitary-type decisions, and while I have not tested this or even know of other people who have tested it, I do not think it has really happened. Maybe people just do not care enough, or maybe people quietly believe that it is good to let the government just run on its own for every once in a while. One version of it could be that people are sophisticated enough to think: sure, I would like more input every day into my government, but I do not want all these other clowns to have input into government every day. It is the other people that I am worried about. I do not know if people are sophisticated enough to be thinking that. They have that potential. But it is noteworthy that it has not happened over the last thirty years. The rise of the internet made it easy to do all this, and yet we have not done it. My default is that if we already got a big shock that should do the thing I said, even getting a bigger shock is not going to change it much.

The fact that the means for very specific, concrete voter action, and the entire system it can engender, have existed for a very long time and we have not done anything with it in that direction yet, probably shows that we care about only a very few select parts of civic life, daily life, governance, and politics in general. So the idea that the issue is with how much democracy we have in the system might be misguided, because it is not about the extent to which a particular system or government-to-citizen interaction is democratic, but more about whether or not the electorate cares about the issue at all in the first place. If it is not something they care about, we see rates of participation plummet. There is a very extreme power law going on here. Does that seem to you to be a threat to the idea behind 10% Less Democracy?

People are a lot more interested in keeping track of sports than they are in keeping track of politics. If you just look around, at least in the rich countries, there are a lot more sports fans who can tell you about the names of different team members than there are people who could keep track of a similar number of politicians. People like you and me who like keeping track of politics — we are the weirdos. People like checking in on their government every once in a while. They keep track of this or that big scandal, maybe, but by and large there are other fun things to think about for most people. There are just a lot of things to think about. The level of checks and balances that we have in the rich democracies — we have a pretty good level of voter participation, and voters are not demanding a lot more. That is worth keeping in mind. The glib version is that people are cognitively lazy, but the more sophisticated version is that they are cognitively economizing. I know the data on this for the US: basically, in the US people start really paying attention to the election about just a couple of months beforehand. They start reading up on the issues, keeping track. Most people are not like us, who just like to know all the time what is going on with this bill or that issue or these political coalitions. One does not want to demand too much of the voters, but the voters are comfortable with not having too much demand placed on them. That is probably the equilibrium for quite some time. That is why you can treat the outcome we have as — one would not say socially optimal in any sense, but it is an equilibrium, and therefore you can use something like the rise of AI and say how does that change this equilibrium. The simplest version is it will help. They will use it to get slightly better answers to questions they were already thinking about, at the input level. It is going to make our voters a little bit better informed, and that is good news for us.

How democratic is the process?

Does it make grammatical sense to say that generally integrating AI into a system or process results in a net increase or decrease in the democraciness of that process?

When I am being precise, when I am in precise mode, I use democracy to mean the effective control of the masses over the government they live in. Giving them more information does give them more effective control. So that is a way in which I would say it would increase democracy. But the more important channel is that it gives them more accurate information about policy, government, whatnot. It will make it harder for them to be fooled. I would not say that that channel by itself makes it more democratic, but it does give them more accurate information and very likely less paranoid information about the world they live in, which will help them make better democratic decisions. There will probably be a bunch of academic research over the next few years about how the rise of AI has or has not improved the information level of voters. My suspicion is that it will improve their information level and reduce the noise level between the true wishes of voters, the deep wishes of voters in some sense, and government outcomes.

My final question on this issue: which does AI impact more, the types and numbers of people who vote on specific things, or the types of decisions that voters make? To be more concrete, we can split up types of decisions voters make to be either the subject matter or domain that they vote — say everyone in this municipality likes to vote on plumbing issues or public-drainage issues but they do not want to vote on economic issues — or whether the decisions are binary, yes/no, or more open-ended. Do you see AI embedding any of these ways of practicing democracy?

My short answer is no. People love a delegate who represents them. For AI to increase the quantity of decisions in a democratic system, if we are talking about the relationship between the voters and the government, it would mean basically having a legislature for plumbing issues, a separate legislature for social issues, a separate legislature for foreign-policy issues. And nobody is going to do that. For whatever reason, the representative-democracy norm is incredibly strong, and we pick one person to sort of represent us on a bunch of stuff, and then that person who is the elected representative outsources a bunch of things to a bunch of little bureaucracies and ministries. That is an equilibrium that I do not see changing anytime soon. There are a few exceptions to the rule, but nobody is crazy about them. California has an insurance commissioner whose job it is to just regulate the insurance industry, and essentially nobody has imitated that. The idea that you have just one person you vote for who makes the insurance decisions has not turned out to be popular. I wrote an essay on this a long time ago, thirty-plus years ago, and that story did not happen. That version of more democracy is totally possible and it does not happen.

Part of the reason it does not work as an equilibrium, at a raw functional level, is that we all know that all government decisions are interrelated. That is a functional, almost utopian reason for why people are not doing this — because it would be bad — and it is a little strange for me to say that as a public-choice researcher, like voters are getting it right, but they are. I expect that equilibrium to stick.

One question is: will the people inside the government start deciding things on a more discrete, minute level? Would you imagine, for instance, that the number of committees in legislatures would increase? That would be a more concrete, more practical version of the situation I just brought up. Again, the problem is that everything is interrelated. At least in the US, the trait in modern legislatures has been the weakening of committees, and more decisions being made at the pure legislature level, because there are grand bargains that have to be made across many committees in order to get anything out the door. That cross-bargaining, and the weakening of committees, has been important in the United States, and without being an expert, without knowing about other countries, I assume something like this is happening in other countries as well. The aggregation of decisions has strengthened over time. Committees are valued for their expertise on subject matter, but they are not given authority on final votes. They have less authority on final votes than they used to. So it is a sort of division of labor: what are the benefits of the division of labor versus what are the benefits of cross-coalition decision?

That centralization is partly about cross-coalition bargaining, and cross-coalition bargainings seem to be really unavoidable, and it seems to have strengthened since 1995, since basically the Gingrich revolution in the US. I do not actually have a great story for why that is true. I could tell you a story, but not one that I believe in strongly enough. My hunch is that AI is not going to change this very much. Cross-coalition bargaining at the aggregate level is probably going to stick with us for a long time, and the AI may be able to offer suggestions to these politicians about some cool angles we could find to bridge the differences across these coalitions. It is a good partner for negotiation. I think that will help. I would expect improvements in negotiation quality due to AI. That is a hopeful piece of news that I think I can offer. Better cross-coalition bargains because you have more people checking out new ideas they would not have thought of on their own, because Claude came up with it.

Ritualized civil war

There is a pretty low-hanging-fruit rationalization for the phenomenon of cross-coalition bargaining. We think that is the happy middle, when it comes to transitions of power, between bloodshed and dictatorship. Instead of one passing it on to his son, or outright revolution and civil war, we have cross-coalition bargaining, and that probably explains its persistence.

One great thing about democracy is that it is ritualized civil war. It is a civil war where we all know which day to show up, and instead of fighting we just count noses and we decide how we are going to move on. Ritualized war is actually really, really excellent. We have social conflict, and to be able to just say we know there are flaws to this mechanism but it is a lot better than the alternative — that is part of how we solve things at the broad social level, and it is partly how we solve things within legislatures. In order to count noses accurately you need to actually have all the noses from the whole legislature and not just one committee.

Exit, voice, and stablecoins

It is interesting that we are talking about ritualized war, because one of the chief sociological functions that ritualized war performs is in disciplining, or at least in performing a public disciplining, a spectacle for people to witness. This is a question regarding the disciplining effects of majoritarian rule versus the disciplining effects of the elite or aristocratic class in any one particular society. A good limit test is central-bank independence, and your arguments about why central-bank independence is a great example of why less democracy is sometimes better. Can we say that for some cases, in the limit, the opposite is true? Say you have a high-inflation country, but because of dollar stablecoins citizens are allowed to exit the national currency without anyone's permission. In this case we have both democracy — people voting through their economic actions; I know this is not the strict technical term of democracy that you use, but it is democratic in nature — and besides that it is also a disciplining device against their own government. If you are not managing monetary policy well enough, or the economy well enough, and our currency is terrible, we want to exit; the economy effectively dollarizes; the government loses control of the entire monetary system. Would this be meaningful to you and your view on what kinds of areas should enjoy less democracy, if we are talking about how to discipline governments?

Here you have hit on one of the great ideas of Albert Hirschman, the great economist. He wrote a book called Exit, Voice, and Loyalty. He said those are the three ways to try to change organizations. Voice is democracy, loosely. Exit is people just leaving the country or leaving the organization. Being able to use stablecoins — say you are a middle-income country with a mediocre central bank, a mediocre fiscal-monetary system — stablecoins give you the power of exit, and that is different from voice. I would say that this kind of exit is not democratic. I would not want to use the word democratic for that. The other term we use is voting with your feet.

If I can put a question on that: does the rise of stablecoins, on net, strengthen or weaken monetary systems in mediocre countries, specifically by way of checking or disciplining against the government? Basically, does the competition make them better or does competition make them worse?

I suspect on net, in the real world right now, it makes them better. The force of competition, the force of exit, actually will make most central banks in the world, even mediocre ones, better than they used to be, partly because they have all seen this same graph created by Alesina and Summers — Larry Summers and the late Alberto Alesina, who himself was a contributor to deep-roots theory and culture-transplant theory. They have this graph showing that countries with more independent central banks tend to have lower inflation, and that it seems to be a free lunch: you do not seem to have worse recessions when you hand monetary power over to the independent bank. This shows up in all of our intermediate macroeconomics textbooks, our monetary-economics textbooks, maybe even a lot of freshman textbooks, and central bankers around the world have learned this lesson and integrated it. The idea that if you have a decently long time horizon, more than beyond say four or five years, low inflation is really good — that is a really powerful norm. Any competition that pushes you to pay more attention to that norm I think will pull them into focusing on low inflation, good monetary systems, even more. That is a practical, cultural, historically informed argument, and I think it is true right now.

In principle you could say the opposite would happen. You could say, well, if everybody is leaving, then the central bank says, what is the point in even trying? The government might just say, what is the point in trying to run our own monetary policy? Let's just all switch to using the US dollar. In principle you could say this could lead to more countries formally dollarizing. I could see a longer time horizon where the rise of stablecoins does lead to dollarization in say four, five, maybe even ten more countries, where they say, wow, it is really easy to dollarize, and the countries that have dollarized have actually done pretty well. They do tend to be small economies. Panama is totally dollarized. El Salvador dollarized. I would say both are true. I will see some more dollarizations as some countries just say, what is the point? Stablecoins have made it super easy for people in our country to use dollars. And at the same time it ends up being a disciplinary effect and making monetary policy better in most countries.

So the net prediction is: the rise of stablecoins ends up meaning we get lower inflation globally, for both reasons. That is a new prediction for me, by the way. I had not thought about this until just now: that stablecoins probably lower the cost of dollarization. If you are a smaller country — if you are a big country, tough, different decision — but if your population is less than twenty or thirty million, dollarization is looking better today than it did ten years ago, because you are looking around and you are like, even Tether works, and now all the big banks are getting ready to run stablecoins, and Stripe is on board, and we all use Stripe. Wow, this dollar is starting to look better.

You heard it here first, just out of the press there.

Agents and crypto

There was a lot of chatter earlier in the heady days of the AI hype cycle about how these two could converge, in that crypto would be the main way that agentic AIs transact in a commercial capacity with one another. Where do you stand on that, and does it change depending on time, or must there be certain conditions under which something like that could play out?

If agentic AI wants to trade, I think the agents want to trade with what is most credible. If they want to engage in high-speed trading — if they are trading stocks at super high velocity — they are just going to use whatever software their companies are using. Those super-high-speed transactions cannot be made with cryptocurrency right now. Those are too slow for high-speed transactions. The big, credible problem of cryptocurrency is that basically writing the code to engage in a transaction just takes too long. Despite the crypto hype of, oh, electronics, computer code is going to be the obvious choice of agentic AI — speed is going to be the first choice of an agentic AI, because they are thinking at the speed of software, and to them Bitcoin transactions are just molasses. That is just so slow. It is going to be way too slow for them. The high-speed transaction platforms that are currently being used in the fintech world for trading stocks are what they are going to use if they are engaging in high-frequency trading. I think that is pretty obvious, and that is going to last for a long time.

Second, for more mundane transactions, if we are letting AI handle B2B transactions for normal supply-chain needs for businesses, those guys are going to be using stablecoins because they need stability. They want to talk to their bosses and show, yeah, I did not engage in some high-risk transaction that took five minutes, in which time Bitcoin dropped 3 percent. They know who their bosses are. They do not want to take that risk.

I do have a hunch. One of my little hunches is that AIs will not just inherit the normal risk aversion of humans. They will actually be loss-averse, in the Kahneman sense of being kind of strangely worried about small infrequent losses. One of the classic explanations for the equity-premium puzzle, why stocks have such a huge return, is that people freak out over infrequent moderate losses to a very strange degree. Even though stocks have a very high rate of return over the long run, people are still very nervous about holding stocks because one year out of every four they drop about 20 percent. Bitcoin is a very volatile asset, obviously. I think AIs will inherit a fair amount of our loss aversion, partly because they are born from us and partly because they know they respond to bosses who are humans. They will be worried about having to tell their boss, oh yeah, I used Bitcoin because it was more efficient and the algorithms are transparent and harder to hack, but one transaction out of every 2,000 I lose 4 percent. They do not want that on their report card.

So agentic AI will like stablecoins when they are doing B2B stuff, and crypto is too slow for true high-frequency transactions.

What excites you

Final question. What excites you about the future?

Artificial intelligence. Really. I think this is a huge promise. There are obvious risks. But the possibility of automating research and development is really what excites me about the prospect of AI: that it can move the frontier forward. Chad Jones, Paul Romer — Nobel laureates; well, Paul is the Nobel laureate who invented endogenous growth theory; Chad, the Stanford professor who just went off to Anthropic on leave — these guys basically built out the theory about how R&D and human economic growth comes from R&D research. Sometimes it is intentional, sometimes it is accidental, but one way or the other we invent new ideas. So far, the prospect of getting artificial intelligences to be supervising, managing R&D labs: that is the prospect that can really double, triple, quintuple annual rates of growth, potentially. The idea of AI automating this job or that job — those have static efficiency gains. The idea of getting AIs to run the research-and-development process, to automate the craft of learning: that is where we get massive economic-growth changes. It will take time for that to roll out, because that is probably an O-ring process with a lot of bottlenecks along the way. But that is where you get the prospect of super-fast growth: from integrating AI into the process of expanding knowledge, not just the prospect of automating existing jobs. That is the world that gets us 10x richer over every twenty, thirty years.

What would you say is the limiting factor for a breakthrough in the economics discipline, maybe insofar as academia goes? How is it similar or dissimilar to the limiting rates in other disciplines? What would be the technological constraint, if any, for the economics profession that AI could potentially be useful?

It is worth noting that economics journals have not been hit by a slop apocalypse, where AI-generated papers start overwhelming journals. Maybe that is happening at the weakest journals, but the elite journals have not seen a big increase in submissions of a bunch of obviously AI-driven mediocre papers. So the limiting factor so far is basically good ideas, and economists know that. The clever idea, the cool insight, the clever way of addressing cause and effect: that is the limiting factor for really great ideas. AI is going to make it easy for us to get access to data sets and whatnot, and for people to double- and triple-check results, but the limiting factor is good new ideas.

I suspect that this will be something where there is a generational, maybe half-generational, change. I could say maybe a generation is five years now. Basically you need a new wave of undergraduates and beginning graduate students who are exposed to the AI method of experimentation, of toying around with ideas. That is when that change happens. That is when we get big payoffs in good economic research, as people who have half a decade of experience toying around with AIs as clever people to talk with, as test beds for new concepts. That is going to really be what speeds things up: having a wave of people who have flexible minds. That will mostly be young. It will basically be young people plus Tyler Cowen who are good at taking advantage of this new tool, people with pliable, flexible minds. That is when we start getting really, really good payoffs. I really look forward — I hope I live long enough to see all these great new payoffs from these people with pliable minds who are toying around with AI to generate clever new hypotheses.

And with that rallying cry, we have come to the end. Thank you, Garett. It has been a real pleasure.

Thanks again for having me. It has been wonderful to be here.