This transcript was generated using AI automation and may contain minor formatting or transcription errors. Please refer to the original audio to verify specific quotes or context.

Ryan Bourne: Frontier artificial intelligence models are improving rapidly, and a new statement signed by hundreds of economists, AI researchers and technology industry figures says the economic consequences could be enormous. Signatories including left-wing Nobel Prize winners Paul Krugman and Joseph Stiglitz, through to George Mason University’s Tyler Cowen and people from leading AI companies, agree on three propositions. First, AI may become radically more powerful over the next 10 years. Second, this could bring untold riches, but also large-scale job displacement. And third, that economists and policymakers must act now, to quote them, “to understand the economic effects,” while building the — and quote again — “incentives, guardrails and institutions to steer the technology to complement humans and benefit society.” Now to many, that statement may sound cautious and reasonable. AI may transform the economy, so government should prepare for that — big deal. But phrases such as “act now to understand” and “steer AI” raise some interesting questions. Are we really failing to study AI sufficiently right now? Is mass unemployment its most important danger? And can politicians ever steer an emerging technology without suppressing innovation or protecting the largest existing companies? To discuss this, I’m delighted to be joined by the economist John Cochrane. John is the Rose-Marie and Jack Anderson Senior Fellow at the Hoover Institution, an adjunct scholar here at Cato, and, in addition to his impressive academic work, a prolific blogger under the title The Grumpy Economist, now housed on Substack. So John, welcome to the podcast.

John Cochrane: It’s a pleasure to talk to you again, Ryan.

Ryan Bourne: So John, you’ve critiqued this letter, but before we get into that, I think it’s worth just asking you how much AI has transformed your own work and where you envisage it going, in terms of helping you. Because quite often people’s critiques of others’ predictions about AI seem to be related to their own view of the potential for the technology. And I suspect you actually don’t really disagree with the first part of the letter — that AI has huge potential.

John Cochrane: Yeah, so far I’m a light user because I’m waiting to see which model turns out to be the best. But there’s all those possibilities there. I happen to believe it will — I think AI is, I hope, one of the defining technologies of our age, as the internet was of the 1990s and the steam engine was of the 1830s.

Ryan Bourne: So that’s certainly true. And how would economists define a big effect from AI? Because it seems to me when a lot of technologists talk about it, they talk about kind of growth rates jumping from, you know, two and a half percent a year to 10, 15, 20 percent a year. But I imagine most economists would think a big effect might be adding 0.5 percentage points to growth each year.

John Cochrane: Thank you for that perspective. Long run growth, despite being the most wonderful thing that we’ve experienced, is remarkably stable. It has in US history been, you know, 2 percent, sometimes up to 3, 3.5 percent, down to 2 percent now, down to like 0 percent in Europe. But even the biggest technological changes have not meant 10, 15 percent growth rates. And should that happen, we should be so lucky. The economic challenge for my entire professional career has been stagnating growth — why were we once growing at three and a half, four percent, and now it’s two percent, half a percent in Europe, nothing and so forth. Something comes along and reignites economic growth, economists largely are out in the streets saying, hallelujah.

Ryan Bourne: So we agree that the technology may matter, may have big effects in economic terms. But the letter’s first concrete warning is about resultant large-scale job displacement. Now, it was striking to me, I think, as it was to you, that this was the biggest concern, rather than a range of other potential downsides of AI, such as cyberattacks, fraud, or military misuse. So why do you think all of these economists highlight jobs as the headline concern? And what risks do you think should rank higher than that?

John Cochrane: Yeah, the unknown unknowns are much higher than the ones you mentioned — cybersecurity and so forth. But even those — you know, airplanes were dangerous when they first came out, and we didn’t require regulators to steer, to tell Orville Wright which way to put the flaps on the airplane. Job loss is the non-economist’s first worry about any new technology. But when we look at the economic history of all the innovations we’ve seen so far — Gutenberg invented movable type, heavens, the monks will be out of a job. No, they turned out to find other things to do. The steam engine is invented — oh no, the guys who get the horse wagons going will be out of a job. No, they found other things to do. In 1900 in the US, 70% of Americans worked on the farm. No, the tractor’s invented, the farmers will be out of a job. No, they found other things to do. Every other productivity-enhancing technology, which is what this is — this is a way of making humans more productive. What that does is, yes, some people lose their jobs, but more people get new jobs. And it’s funny, there’s this 1930s mentality, this lump-of-labor fallacy, we call it in economics, that anybody who loses a job is therefore permanently unemployed. I think our economy — the numbers I remember — we, two and a half million people lose their job every month, if I got the number right, and 2.6 million people get a new job every month. There’s a lot of turnover in jobs. First, and second, AI and its productivity is going to create lots and lots of new jobs. And finally, the jobs fallacy in economics — in the end, why does India have a billion jobs and the US only 250 million? Is that because of stimulus? No, they have a billion people. They have very low wages. If you want to worry about something, the issue is wages in the long run, but not jobs. So we’ve seen this movie over and over and over again. Some people will lose their jobs. More people will get new jobs. Some people’s wages will go down. People who know how to use AI, their wages will go up. Our children and grandchildren will be dramatically better off if it pans out the way they say it will.

Ryan Bourne: Yeah, so I think a lot of people who envisage technological unemployment imagine some sort of Big Bang shock. But as you say, what tends to happen over time is that particular trades kind of slowly disappear, or you get the nature of a particular job change as technology augments it. Now, of course, some communities can suffer badly from that when you have rapid technological change, but permanent economy-wide unemployment never seems to have followed. So for those who are non-economists listening to this, you say new technologies destroy jobs, but also create new ones — walk through the basic mechanism for how that happens, because there’s a lot of different ways you could imagine that happening. One being AI improves people’s productivity, that boosts wages, and that creates new demand for other industries. You could have new industries emerge because of the technology itself, facilitating types of commerce that we can’t even envisage yet. So when you talk about the creation of new jobs, what’s the kind of mechanism that you’re thinking about?

John Cochrane: Well, I love what you said, “they can’t even envisage yet,” which is this is prime Cato — the sort of thing we keep reminding people, it’s not take today’s economy and tweak it, it’s all the new businesses and new opportunities that will spring up. Trucks didn’t just substitute for horse wagons — people found new things to do with them. And one of the things people keep saying about AI — I’ll just mention a couple of those coming out of the economics study of this stuff. AI may, or could — it hasn’t really made a big impact on the productivity and employment statistics yet, this is all totally hypothetical. AI may or could shrink the necessary size of a firm, because think of what you can do if the HR, compliance, legal, and all the rest of that stuff is run automatically. Well, then you and I can start a new company making libertarian podcasts, and compete with Cato — I’m simply joking here — but you get the picture. So the opportunity to start lots of new companies is there. Second, even within existing companies, I’ll just mention three channels that people tend to forget. Most jobs are bundles of tasks, and some of those tasks get automated, allowing you to do better on other tasks. If I could only get AI to answer my email, then I could double the work hours I can do on other things, and that sort of thing I think we look forward to. And a big one: output expands. Every time existing workers become more productive, that means the output of the company is cheaper to make, and competition means they sell more, and the output expands dramatically, even given the same number of workers — exactly what we’re looking for, together with the output expanding of the new industries and so forth. The prime one people study is the textiles that came into the United Kingdom in the early 1800s. And yeah, the weavers were unhappy — the high-skilled weavers that competed with the automated looms, they lost wages. But jobs increased dramatically, and people moved off the farm into factories. Now, those were dark, satanic mills, dangerous, long hours and all the rest of it, but they were a hell of a lot better than being an agricultural laborer in 1800s England. And the output of textiles dramatically increased, so everybody got cheap clothes. And that’s the sort of thing that can happen again.

Ryan Bourne: So I guess if I was kind of steelmanning the proposition put forward by the signatories of the letter, they would say, look, okay, there’s not this aggregate job loss, job displacement so far, but some of your colleagues at Stanford are trying to study and follow the data on unemployment effects across the labor market. And they’ve identified an apparent decline among 22- to 25-year-olds — entry-level age workers in the occupations most exposed to AI. Now, of course, that doesn’t prove AI caused that. But do you see that as a kind of possible canary in the coal mine? Does it make you pause on the confidence that this will be the same as those historical trends of new technologies simply changing the nature of jobs? Do you think there’s a large possibility that this time might be different?

John Cochrane: There’s always a possibility, but we don’t know. And you set out the letter very clearly in your opening remarks, but not with quite the force and outrage. Let me just review where we are. AI might have big changes. It could result in some job losses. We must act now to understand it — as if every single economist isn’t spending their whole lives doing that. Right now, the outpouring of papers on what might AI do is enormous. And it’s kind of funny — the outfit that put this together, the Stanford Digital Economy Lab, does nothing but produce papers on AI. In fact, the one place that AI has really increased productivity is in writing blathering papers about what the future of AI might bring — and now ChatGPT can write those for free. We should really see an enormous increase in the number of academic articles about what AI might do, none of which is particularly grounded in reality. And admitting in the first sentence, “we don’t understand what the heck it’s going to do, so subsidize our research,” but “we must act now to steer it” while we don’t know what the heck it’s going to do — that’s what sparked the grumpy economist to get grumpy. But finally, to answer your question: there’s some anecdotal evidence that 22- to 25-year-olds are having a harder time finding jobs. Of course, that comes and goes — there’s waves of ease or unease in college employment. I would suggest that a lot of college graduates are extremely badly prepared for a tech world. If you got your degree in communications and gender studies, maybe you weren’t exactly prepared for any workplace, let alone the AI workplace. And you’re not going to get a $500,000-a-year job working for a social justice non-profit and live on the Upper West Side of Manhattan. That’s not the fault of AI, that’s the fault of a whole lot of other things going on. But there is that little statistic, and there are other countervailing statistics — we are in an unbelievably healthy labor market. The US economy is trundling along with a 4.1% unemployment rate, something not seen — it’s been that way for a couple of years now, but not seen since the 1960s. Job openings remain strong. So you see AI everywhere except in the productivity statistics and the labor force statistics, given the kind of apocalyptic “we must steer” sort of language. We’re not even beginning to see anything on a factual basis. And there’s no need to act now before you see it, unless what you want to do is stop us from this wonderful technology because you’re scared of it. One can react to these things in real time if and when they occur, if what you want to do is improve labor markets and nonetheless have the benefits of the technology. You have to wait and see.

Ryan Bourne: So as you say, there’s kind of two calls to arms within the letter. One is the “we must act now to understand,” which is about encouraging economists to research. But as you say, a lot of economists, think tanks, whatever, are already doing a hell of a lot of research on AI. You kind of alluded to it there, but did you see that explicitly as a kind of opening to demand more government research funding, or did you just think it was to urge other economists to focus on this issue instead of studying inflation or game theory or whatever else they’re doing?

John Cochrane: Well, coming from a digital economy lab at Stanford, certainly it sounds like a call for “we need more subsidies for our research,” or support for the research, or “come join us” — but there’s hardly a need. There’s supply and demand of research as well. The fact is, studying the possible effect of a new technology — economics is just not well adapted to doing that. We can look at historical precedents, so a reminder of how the last 500 technical innovations have gone is useful, and doesn’t tell you the kind of apocalypse you’re looking for when in step B you want to run to Washington and say, legislate things. As you said, it’ll take a long time. Productivity will increase slowly. People will adapt. It’ll be wonderful in the way it will be met — as we met airline safety, along the way there are problems and you figure them out and you fix them. So that kind of research is there. But my point was supply and demand of research — spending a lot more time on things that nobody knows and nobody can know is not, in my view, necessarily a good idea. But I’m a free marketer — let supply and demand work. Where’s the market failure that is stopping research on AI from going forward? Economists are terrible at understanding completely new technologies. Look at the last great fashion in economics: to estimate what the effects of climate change on the economy was going to be — what’s going to happen 100 years from now, depending on the level of carbon in the atmosphere. The research on that was, how can I say politely, extraordinarily low quality, because it’s very hard to know what’s going to happen 100 years from now. So a totally new technology that we don’t know what it’s going to do — you can kind of repeat the old labor market stuff, which people have done, and started to think about what happens if new tasks are related. But a vast effort of research in that doesn’t seem to be very well placed.

Ryan Bourne: Yeah, I mean, even five or six years ago, I remember a lot of economists and lawyers spending vast amounts of time debating some of the potential downsides of the way the technology and social media ecosystem was evolving, and talking about various regulations and utility-style regulations of those platforms to try and encourage a bit more static efficiency, a bit more consumer surplus. And of course, then AI comes along and completely disrupts that whole sector and changes the nature of those technologies too. So a lot of this feeds into sort of the catastrophism that our public policy chattering classes like to indulge in.

John Cochrane: And I’m glad you brought us back. Yes, it was just a few years ago that Twitter was going to rot everybody’s brains and we need regulation to do something about it. And, you know, this is in some sense a mind of — oh, the population bomb is going to come, we need to do something about it. Oh, the resources are going to run out, we need to do something about it. Oh, nuclear power is going to turn all the frogs blue, we need to do something about it. The climate catastrophe is coming, we need to do something about it. So every five years, you need a new “just over the horizon” — nobody can really see it now, but subsidize us to study it and regulate it ahead of time. That seems very popular.

Ryan Bourne: So obviously, as libertarians, we worry about the government trying to steer a new technology, and that language was striking. But I think it was particularly worrisome when it was combined with the end of that sentence — “steer AI so that it complements people and benefits society.” Now, that might sound benign to the untrained ear, who doesn’t want a new technology that benefits humanity or whatever, but the “complements” language combined with the job displacement kind of implied that perhaps we should be trying to — or government should be trying to — steer the technology to avoid the displacement. And over the long run, a lot of those productivity gains come from automation, from new technologies replacing human labor. So I saw that as particularly worrisome. Did you interpret it in the same way, or am I overinterpreting that?

John Cochrane: Absolutely. As libertarians, we need to avoid this common assumption that people make that everything needs to be regulated and governments are able and good at steering things where we go. Remember, when you see “benefits society,” any American who knows anything about the philosophy of our government, as well as a libertarian, says: who decides what is the benefit to society? Is it people ponying up their own money to say, “oh, that benefits me”? Or is it a government bureaucracy quickly captured by some political movement who decides what the benefit and equity and all the rest of it is? So that’s the real danger. Steering new technology — there is a danger. Most of the time, what government does is slow down economic growth, slow down progress in order to protect somebody or other — usually somebody politically powerful, but sometimes people with less political power. But that’s what it does. Uber comes in and wants to stop the taxis, Waymo comes in and wants to stop Uber, in order to protect the Uber drivers. And there’s a real danger here that, in order to protect current rents, current ways of doing things, we simply slow down this technology, which is the cat’s out of the bag here — we can slow it down in the US, but we’re not going to slow it down in China. And I think that wonderful competition between countries is one of the best things for keeping us from simply turning off the technology in order to protect old ways of doing things. But that happens. The famous example of the Chinese treasure ships — China in the 1400s had ships better than anything the Europeans had. They decided it was socially unstable, so they burned the ships and the plans, and that’s why you and I speak English and not Chinese now. And more recently, Europe has turned off genetically modified foods — for a generation, they didn’t get the benefits of that. They’re trying to turn them off in the developing world so that kids can’t get vitamin-A-enhanced rice and so forth. Nuclear power got shut off on sort of vague fears of things that might happen, for a generation, and that’s why we have a climate problem, and that’s why we have expensive electricity. So it’s very possible for vague fears to be the public face of rent-seeking protection, which is what the government largely does. It’s very hard for me to think of an example where the government steered a new technology in a way that really got the innovation part of it, the economic part of it, going — as opposed to steering a new technology basically to slow it down and protect somebody or other.

Ryan Bourne: Well, I guess somebody might say stuff like the advanced market purchase orders for the mRNA vaccines or things like that. But they’re few and far between.

John Cochrane: There are a few, and that was a great one. Hilariously, then, of course, the Republicans changed their mind and decided vaccines were bad, and the Democrats changed their mind and decided vaccines are good, the minute the election changed. And the government has in the past done a good job on some basic research, so that’s an important part of our ecosystem — though an entire other podcast should go into how the basic research ecosystem, any monopoly, is always in danger, and that one is kind of fraying at the edges. But the applied research — the internet may have been, the HTTP protocol may have been invented as a side project at a physics lab, but turning it into something useful like Zoom was entirely a private-sector effort.

Ryan Bourne: So the letter itself is vague about what guardrails and institutions would mean in terms of steering the technology, and I’m sure a lot of the signatories have different ideas, but at least two of them — I’m not going to say the specific names because I don’t want to personalize this — back regimes under which frontier AI models, or the companies developing them, would have to convince government regulators that a model was sufficiently safe before releasing it. So this is basically like a Food and Drug Administration — what that does for drugs — an equivalent regulator for AI. So what could go wrong, John?

John Cochrane: Well, let me sound off on three aspects. First of all, I’m a little embarrassed for my fellow economists who signed this letter. Economics has a framework for thinking about regulation — you and I disparaged the framework, but let’s bring back the framework. The framework is: there’s sort of a presumption that markets work, but if you want to regulate, document a market failure, show how a regulation could fix the market failure, and hopefully also think about regulatory failures and how you’re not going to repeat those. But where’s the market failure? Nobody’s said that, other than “it could do something.” So at least bringing back that perspective would be useful. And a preemptive permissions regime frightens me in many ways. Take one thing that AI does that we’re all familiar with — ChatGPT and Claude and so forth, the chatbots that are all used. This is a way of spreading human knowledge from some humans to other humans, because it’s trained on human inputs, and it allows you to access what the smartest people around the world know very, very quickly. But that’s communication, and it’s catnip for censorship. And that’s what we’re already seeing — try to ask any of the models about anything vaguely sensitive. If you ask them about vaccines, they will shut down quickly and give you the party line on vaccines. And so if you have to submit a model to a regulator ahead of time to say “is this safe,” you know exactly what the regulator is going to do. The first thing they’re going to type in is “tell me who really won the 2020 election” or “tell me about Tiananmen Square.” We know the Chinese already do that, they try to stop it. Now we’re back into the usual cat and mouse of trying to figure out how to prompt it to get it to tell you. But the number one thing governments do when they’re asked to preemptively regulate a means of communication is censor what’s in there. So that’s the obvious, screaming danger of the thing.

Ryan Bourne: So economists’ letters and statements don’t have a great track record. But I think it was noticeable that some of the signatories are not the usual suspects for these things. And others were actually people at companies like Anthropic, OpenAI and Google. So I guess some will look at this as a sort of “Nixon goes to China” thing, saying, see, even these market-leaning economists and the companies themselves developing these products recognize the potential devastating effects on labor markets. But of course, as good libertarians, we recognize that certain regulatory environments — especially government approval of AI models, which we just spoke about — help close off competition and entry to the benefit of existing companies. So maybe I’m being a bit too negative on that. But I’m just trying to think why representatives of these companies in particular would have been willing to sign this. That’s an obvious justification, but perhaps I’m being too down on them.

John Cochrane: Yeah, this is the problem we’ve always had with regulation. The bankers know more about the banking industry than we do, so they come in saying, “oh, we need regulations because of X, Y and Z” — and gee, it always ends up helping the banking industry. So one can be a little bit cynical. First of all, these are companies who aren’t making any money and want to attract investors, so saying “AI will be incredibly powerful” is kind of helpful talk in their book. Second, they’ve invested hundreds of billions of dollars in AI models and data centers, which we recognize quickly as a fixed cost — a large fixed cost, low-marginal-cost business. If you’ve just put $100 billion into an AI model, you’ve created something that has a half-life of about six months and no profits. So the last thing you want is somebody else coming in and building on your idea, and then you lose the money. So the demand for regulation from companies has always been about protecting themselves from competition, especially in these early stages where these fixed costs are enormous and the profit stream is really not sure enough. I am a little bit skeptical. Also, people who are really good at training neural networks are not necessarily really good at understanding competitive labor markets. So while they may know a lot about where AI might be headed in the next six months to a year, they have no special expertise in understanding what’s going to happen to jobs — whether the auto repair shop that I just went to, trying to use its AI assistant to schedule my appointment, whether that’s going to work out or not, and whether they’re going to fire people based on that result or expand.

Ryan Bourne: So the letter doesn’t tell us exactly what government action would be. But in California, Gavin Newsom’s recent executive order outlines one concrete vision of this. As I understand it, he’s ordered his government to review the best evidence on early warning signals on labor market disruptions, and then advocated a bunch of potential policies to deal with any downside. So what exactly is he proposing?

John Cochrane: So when I went and read Newsom’s AI order — Trump also has an AI order that’s probably worth reading too, but this, I thought, was a useful insight into where the sort of centrist wing of the Democratic Party would be thinking about their AI efforts when, in the inevitable turn of affairs, they get their chance to govern by executive order the same way we do. And it was a striking order. Again, it was entirely concerned with the potential labor market effects. And the solutions he wanted to look at, I found hilarious, because they were straight from the 1930s: worker-owned collectives — that’s the answer. Subsidized employment. Yes, indeed — please, what’s going on here? And support for downtrodden areas — Fresno has an unemployment problem, it has nothing to do with AI. And there, I think, there’s a big danger: the cure is worse than the disease. We have big labor market problems in the US. One of the big reasons we have people who find it hard to get jobs is because it’s hard to fire people. So labor market protections are particularly dangerous in an era of labor market turmoil, because the best way to guarantee that you don’t have mass unemployment is for people who lose their jobs to AI to easily be able to find new jobs. But nobody hires new people if it’s difficult to fire the old people. This is something that’s really hurt Europe — it’s very hard to fire people in Europe, it costs $100,000 to fire somebody, so they don’t hire people unless they know they want to keep them for the rest of their lives. So the main answer to mass unemployment is the ability to “take this job and shove it, I don’t work here anymore” — which means you can get a job somewhere else. So if you’re worried about unemployment, and employment is kind of a catastrophe in California, people are leaving the state, and especially people with modest means are finding it hard to work here, well, getting out of the way might help. But more of that disease is likely to be where the Gavin Newsom-to-the-world approach ends up.

Ryan Bourne: So I think one thing that worries me about the letter — the call for steering the tech in general — is that it reflects people getting ahead of themselves in proposing policy for eventualities that we just haven’t seen yet. You mentioned that earlier.

John Cochrane: Well, if I may interrupt — yeah, so what struck me about Gavin Newsom is that these were the same old answers in search of a new question. Worker-owned collectives, we’ve been after this for 70 years. Oh, AI, that’s the reason now for worker-owned collectives. Come on. Go ahead, I’m sorry.

Ryan Bourne: Yeah, I mean, in general, this type of thinking could have damaging consequences in other areas. So just think through a couple of examples: Congress might decide not to do anything about the debt path because they hope AI-fueled growth will bail them out. Now, not only do we not know whether or how much AI will actually boost growth, but there are lots of complex links between growth and the federal budget. If AI’s primary benefit is that it improves our outlook for health, diagnostics, and improves longevity massively, we might actually see aspects of the federal budget worsen, given our entitlement system shells out money to the elderly. So I think that’s an area where presuming that AI is going to lead to a huge takeoff, as a means of not doing anything, might be dangerous. But that same speculation is feeding through to advocacy in other areas. I know Elon Musk thinks there’s going to be this big productivity boom, huge abundance, and his proposed remedy to a lot of people being displaced in the labor market is universal high income. But area after area, people are talking about these things before we’ve seen the high growth, let alone the job displacement.

John Cochrane: Yeah, there’s spending the goose that lays the golden eggs — we’re spending the golden eggs before we’ve even bought the goose, as far as I can tell. Higher productivity growth would be very useful for the federal budget, though it has to translate into taxable incomes to actually massively increase revenue. And you’re right, Congress also has a great ability to spend whatever boom comes in. So I think counting on AI to reverse a half a century of slowly declining growth and slowly worsening federal budgets within a short timeframe is a very poor kind of planning.

Ryan Bourne: It would be really remiss of me, having you on, not to ask you a question about macroeconomic policy. So I think a version of that same optimism has also entered some of the monetary policy discussions. I know Kevin Warsh, the new Federal Reserve chair, prior to taking up his position, implied on a couple of occasions that he foresaw this huge potential AI boom to productivity — that AI could put such downward pressure on prices that it could facilitate lower interest rates. But it’s not obvious to me why a much higher growth world would lead to lower interest rates. So I wondered how you interpret what he’s trying to say about that, and whether you agree with him.

John Cochrane: Well, some of it was undoubtedly a need to square a circle. If you’re in Washington right now, you want a reason for lowering interest rates. What could we possibly come up with, given that inflation remains 3% and has not gotten back under control? Oh, productivity will come and drive down prices of everything, so we need to get ahead of that. Again, not obvious why you need to get ahead of that, even if you could forecast that it was coming. But I think the case is less clear. First of all, from a monetary point of view, from the economic growth point of view — all of a sudden your wildest dreams come true, and this is a problem. A big productivity increase that lowers the price of things is the negative of a supply shock. The usual conundrum is stagflation — prices going up and employment going down. A boom with lowering prices is sort of a central banker’s dream — you get to stimulate the economy and have low inflation. So don’t count on paradise coming before you really see it. Let’s just think through the basic economics. Suppose you even knew that productivity was going to start skyrocketing, which it is not doing yet, and it always takes a lot longer than you think — but suppose you knew that was coming. Do you need to get ahead of it? What it does is make the prices of goods and services go lower relative to wages. Well, with voters screaming about affordability, what is so terrible about prices going down relative to wages? Deflation is usually thought to be bad, but in the 1930s, deflation came from what’s called lack of demand associated with very high unemployment. Deflation in a boom, because the factories are humming and churning things out cheaper and cheaper, means everybody’s wages go further. Why is that a problem that needs fighting at the Fed in the first place is a good question. What we’re talking about is letting the Fed inflate in a way so that wages go up at constant prices, rather than prices going down. It’s not obvious why you have to do that in the first place, rather than say, “hey, we cured the affordability problem, thank you.” And it’s not obvious that interest rates need to go down. There’s supply and demand for everything. Remember, what we do at Cato is remind everyone, there’s always supply and demand. If AI is the great new thing, then it needs big investment — AI data centers, AI models and so forth. Big investment means you need to attract investment funds, that needs higher rates of return on investment, the marginal product of capital is higher. We need to encourage people to defer consumption until the great new stuff comes. So there’s the famous R‑star — the real interest rate is arguably higher in a time when there’s a big productivity boom needing investment in physical capital. So all around, I’m a little dubious of the argument that we should lower nominal interest rates right now aggressively because AI will create lots of productivity unseen in the last 50, 60 years of growth, and we need to get ahead of it.

Ryan Bourne: It’s probably a discussion for another day, but it’s evidence, I think, of my worry about an inflation bias creeping into Washington. Whenever there’s, as you say, a negative supply shock — oil prices go up — people say the Federal Reserve can’t alter the global supply of oil, we shouldn’t do anything, we should sit on our hands and just accept the price level is going to go a bit higher in the short term. But then when you hear about a potential positive supply shock in terms of faster economic growth, faster productivity, a lot of people pivot instantly to saying, well, this will allow us to loosen monetary policy — and now we really care about hitting our inflation targets strictly. So there’s an asymmetry there. But I’m afraid that’s all we’ve got time for today, John. We’ve really appreciated your dissent on this issue. The political reflex, of course, is to assume that every powerful new technology needs a bespoke framework before anyone quite knows what it does. We, as libertarians, begin with the opposite presumption — not because new technologies create no problems, but because experience tells us that heavy-handed rules can choke off innovation and entrench today’s incumbents. So we think AI will undoubtedly create new challenges, some of which we can’t even foresee yet. But I think far too much of this debate has already been consumed by confident predictions about the future — confident predictions of mass technological unemployment — when the evidence remains thin and history gives us good reason for skepticism.

So thank you for listening to today’s episode of the Cato podcast. Once again, I’m Ryan Bourne, joined today by John Cochrane. If you enjoyed today’s discussion, please subscribe and leave a review wherever you get your podcasts. To learn more about the ideas and research discussed in this episode, you can visit Cato​.org. The Cato podcast is a production of the Cato Institute, dedicated to advancing individual liberty, limited government, free markets and peace. Join us next time for more insights and conversations on the issues shaping our world.