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.

Adam Michel: Anxiety over AI has created a new topic that Republicans and Democrats can agree on: we should tax it. And with most bipartisan agreement, this consensus is a bad one. Specific proposals vary; Andrew Yang wants to tax computing power, Bill Gates wants the robots to pay taxes, Elizabeth Warren thinks her wealth tax is the solution, and Bernie Sanders wants taxpayers to own about half of AI companies. Donald Trump’s view doesn’t seem that much different. The proposals all vary in their mechanisms, but they are motivated by the same three-part story. First, that labor’s share of national income has declined in recent decades. Second, that AI will accelerate this loss of economic power for workers. And third, that the tax code makes all of this worse because we are already undertaxing capital.

The problem with this is that all three parts of this story are wrong. I’m Adam Michel, Director of Tax Policy Studies at the Cato Institute, and my guest today is Daniel Bunn, President of the Tax Foundation. We’ve each written on this topic recently and came to similar conclusions that this AI panic is generating some genuinely terrible tax ideas. So today, we’re going to take them apart one at a time. Daniel, welcome to the Cato Podcast.

Daniel Bunn: Thank you for having me.

Adam Michel: Before we get into the specific proposals, I want to start with a big picture question. What do you think is driving this AI anxiety, and what are the things that made you decide that the reactions that we’re seeing needed a response?

Daniel Bunn: Yeah, the biggest thing that I think is driving this issue is just the uncertainty that surrounds what this technology will mean for the economy. And that could be good things that it could mean for the economy or some negative things that some people have been talking about. And that uncertainty is not something that individuals and families like living with. Policymakers in some cases like to use that uncertainty to drive a certain policy agenda, maybe a pre-existing policy agenda, saying, “Well, now is the time for this solution that I’ve always cared about,” just because people are uncertain about a new wave of technology or a new wave of change.

Adam Michel: Great. So let’s start with that underlying problem that seems to be driving a lot of the anxiety that we’re seeing. Because if the diagnosis of what’s going on under the surface is wrong, I think the prescriptions will also be wrong. The frame that I see that everyone’s using is this sort of capital versus labor, AI versus workers framework. It’s a zero-sum fight over some fixed pie; I see it everywhere. Sam Altman, CEO of OpenAI, worries that AI will, quote, “break capitalism” by shifting more leverage from labor to capital. This is just an updated version of something that Scott Bessent said before he was Treasury Secretary, making a similar point that workers have been sort of systematically disadvantaged relative to capital since the 1980s. Can you give us a bit of a history lesson? Have workers been systematically losing over the last several decades?

Daniel Bunn: Yeah, this is something that comes up all the time in economic policy, people pointing to different measurements and saying, “Well, it looks like the shareholders or those who are more well-off are earning a lot more or getting a larger share of income from their work or from their investments.” But the truth is a lot of those statistics don’t actually net out in the way that you should net things out in order to get true measurements of who is deriving value from overall economic activity. So if you do things the right way, for capital income, you have to look at net income. You have to look at the income minus the depreciation that happens in the economy, minus the things that don’t really show up in people’s pockets. And when you do that, you get something that’s kind of fascinating in economics: you get a trend that’s really just a flat line. Actually, it’s two flat lines. The labor share of national income has been roughly around 70% for almost 100 years. Basically, since the time we started collecting some of these economic statistics, you can see the stability in the labor share. And since these things add up to roughly 100%, right, unless you have some measurement error, you’ve got 70% going to the labor share and 30% going to the capital share.

That to me shows that through this last century, with all the different technological and geopolitical changes that we’ve seen in the world and in the economy, these are relatively stable relationships over time. It also makes me think that these terms the debate often gets couched in—of capital versus labor—ignores just the working reality for most people. It’s a capital-labor partnership rather than some sort of oppositional thing. If you and I didn’t have computers, maybe we’d go on a street corner, I guess. But the partnership between capital and labor is something that’s so pervasive throughout the economy and has been part of this productivity and growth boom that we’ve seen over the last 100 years as well.

Adam Michel: Yeah, I think that when you just look at the chart, it is remarkably stable. This idea of the share that workers are benefiting from, the share that workers are getting from this broader economic output, you see something similar on the wage side of the story as well. There’s this pervasive misunderstanding that wages are stagnant when, when you properly adjust for inflation, wages are up quite significantly over the last several decades. The sort of cost of living has actually gone down by many measures if you take into consideration the quality of things changing throughout the economy and how many hours of work you have to put in to buy a certain basket of goods. And so by many different measures, I think it’s pretty clear that workers are not being systematically pushed out of the economic pie. Another piece to this is a lot of times people will point to how well the stock market is doing. Yes, the stock market’s doing great, but the great thing, especially in the US, is a lot of people have a large portion of their savings exposed to that stock market performance as well. So even if people want to try to separate capital versus labor, a lot of workers, whether through pension plans or 401Ks, are exposed to the benefits from the stock market and kind of the capital income side too.

Adam Michel: Yeah, it’s certainly not a clean story. Even if you don’t have a 401K, your teacher pension or your firefighter pension is also invested in the stock market, and so there are lots of ways that people benefit from even the capital share in that equation. You alluded to this in the latter part of your answer there on the complementarity between capital and labor, this idea that it’s not a fight between machines and workers. The historical record pretty clearly shows that, more often than not, new capital investment, new technology, and new machines work with labor to make them more productive, to allow them to earn higher wages. I know this is a core feature of a lot of the work that the Tax Foundation does. You have a model where this is like the drive, the engine inside of it. Can you just elaborate on this relationship a little bit more? How do capital and workers work together?

Daniel Bunn: Yeah, so there are two real conceptual pieces that are helpful here. Certainly, there’s a lot in the academic economics literature over the years that shows this relationship between capital and labor, and investment in capital and investment in human capital that drive growth over time. But I think a lot of people focus on the “maybe a robot took my job” side of capital versus labor, rather than seeing that if something is being automated or simplified through labor-saving technology, it frees up hours of work for higher-value work for individuals, or creates an opportunity for another entrepreneurial business model. There are also plenty of jobs where capital and labor are obviously working together really well, whether it’s being able to automate a part of a process so you’re able to spend your time serving more clients or larger markets, or simplifying things to the extent that the risks that might have been associated with a job 10 or 20 years ago because it was simply dangerous are mitigated. Technologies have come online so that those jobs are much safer and more productive over the course of time. This is really the type of thing that we like to see out in the real world, but it’s underpinned by decades of economic studies on the value of capital investment and where that is complementary to workforce expansion and workforce investment. That ties right into other parts of the economic literature about human capital formation and skill building, where people can move from lower-skilled work to higher-skilled work or a higher level of innovation or activity. I think one of the promises of AI specifically is that it can lower the barriers to doing more complex tasks, so there’s a whole new way in which this technology can help people sort of move up the economic ladder.

Adam Michel: If workers and their tools are complements, as you’re saying, which I think is correct, then taxing the tools to help workers is a strange way to go about helping them. But before we get to AI specifically at a high level, tell us a little bit about how taxes on capital are a problem in the first place, and who ultimately ends up paying these sort of broad taxes that are trying to hit capital in some way. By capital taxes, I mean things like the capital gains tax, taxes on businesses, and the corporate income tax. These are generally things that economists think are capital taxes, right? They’re falling not on labor, but on capital.

Daniel Bunn: Yeah, a lot of people try to separate out the capital gains tax versus your payroll tax or your income tax. Those corporate taxes or capital gains taxes, a lot of people think, “Well, yeah, that’s taxes on other people or on the companies.” But it really comes down to the burden on shareholders, the burden on workers, and the burden on consumers. Depending on where you are in the business cycle or in the market, or how quickly you’re able to adjust to different margins of taxation, sometimes—if there’s this theoretical idea of a temporary, one-time tax—that’s just going to fall on the shareholders, and the shareholders are going to be able to adjust their investments and bear the burden of a capital tax. But in reality, businesses are planning for the long term, and they’re planning hiring decisions and investment decisions. If you increase capital taxes, you’re going to impact the ability of companies to invest and hire, and those are going to fall on individual workers. Companies might also, in some cases, be faced with a tax that they can clearly pass on to their consumers through higher prices.

So these are all elements that get to the challenges with capital taxation—challenges and risks, honestly. A lot of times, what policymakers theoretically want to do, or what we would like them to want to do, is to raise revenue in a way that avoids distorting economic decisions. Capital taxation is one of the most distortive of economic decisions. You want to let those investors, business leaders, or the people who are planning expansions into new markets or into new communities make those decisions and minimize the tax wedge on those decisions as much as possible. Because if you don’t, then there are fewer job opportunities, fewer business expansions, and potentially price pressure on consumers.

And so it really goes back to the story we were just talking about before: this idea that the way workers get more productive and can command a higher wage is that they need to work with newer tools and better tools. If a tax means that someone’s going to do less investment—whether the tax is on the business making the investment or on the capital gain that is incentivizing someone to invest in that space—if you’re just getting less investment, workers are getting less of the thing that will make them more productive, which ultimately means they can command a lower wage. This is how taxes on capital often get shifted to workers in the form of lower wages. It’s sort of this backdoor way of actually hurting workers when politicians think that they’re taxing the big titans on Wall Street or something else.

Adam Michel: So now let’s apply some of this to AI specifically. The features that everyone’s panicking about—that capital is incredibly mobile, that it substitutes for labor really easily, that it’s sort of this infrastructure that can build itself—these are exactly the features that make a tax on capital shift hardest to workers. So a compute tax, I don’t think, is an exception to this rule that we were just talking about; it seems almost like the worst case of it to me. Can you take us through some of the actual proposals of AI taxes and some of the problems that you see with them?

Daniel Bunn: For any tax, you need at least a tax base and a tax rate. People throw out this idea of a compute tax as if they know what base and rate to use. To me, it’s possible, like you mentioned earlier, Adam, that it’s strictly a tax on computing power—that you’re taxing the energy used by data centers or somehow taxing the tokens that are generated as part of using these AI tools. That’s more of a token tax than a compute tax, but it’s all in the sense of, “We’re going to identify this use or this piece of the AI puzzle and we’re going to tax that.” Now, it doesn’t tell me anything about what amount of money they’re trying to raise, and that’s where really the rate comes in.

You need to be able to measure these things. I did some brief overviews trying to understand what the base would be if we were taxing compute, and it’s really hard to differentiate from the statistics that we have between energy for AI versus energy for non-AI. It’s almost impossible. Tokens are a little easier to measure; that’s more of a question of like a sales tax on products being sold. But then if those tokens are being sold to businesses, I don’t want taxes on the inputs to other businesses that are using the AI tools. The problem is there is no definition of what compute is, but it’s also manipulatable, right? As soon as you put a tax on one of these things, they’ll change how they do it. And so the base is what economists call elastic.

Adam Michel: Yeah, the compute tax is certainly one that’s been talked about. There are also proposals or ideas to align the taxation of capital and labor. Elizabeth Warren thinks that now is the time, if ever there was a time, for a wealth tax because of the value that some of these companies are seeing in the markets or through the different models and releases driving up value for investment. There are also calls for tax cuts, like cutting taxes for workers because workers are going to be harmed, or maybe we need this sovereign wealth fund that is somehow a stock-picking exercise, I guess, to fund—

Adam Michel: That’s what I want my politicians doing.

Daniel Bunn: Yeah, exactly. To fund this maybe semi-endowment for a universal basic income for harmed workers. All of these are, I think, in a lot of ways missing what I would want people to think through when they come to a tax policy conversation. At the Tax Foundation, we talk about our principles: simplicity, transparency, stability, and neutrality. Almost everything that I just mentioned ignores at least one, if not all, of those principles. To the point that you’re making about a compute tax, you’re essentially saying, “Oh, man, there’s this really new innovative sector. Let’s tax the inputs to that sector.” There are certainly taxes that AI companies pay and will pay; whenever some of these companies become profitable, they’ll pay property taxes—I think we’ll talk about that a little bit later—and they’ll pay whatever energy taxes they’re liable for. But one of the key lessons in tax policy for businesses is that you want to avoid taxing those inputs. If I’m a consumer and I’m buying food for myself, that’s me as the end user. But businesses are in the business of production, and you don’t want to tax inputs to that production. Otherwise, you’re throwing sand in the gears of whatever that business is trying to build and sell. At this point where we are in the AI cycle, I don’t really want anybody throwing sand in the gears. I kind of want to see where this thing goes. It’s likely or possible that some of these business models are going to fail, but that’s kind of the way our economy is supposed to work.

Adam Michel: One of the other things that’s going on here is all of these definitions that we’ve talked about and throwing sand in the gears of these systems. These companies are incredibly innovative; they’re at the forefront of their technology, and trying to tax either their inputs or their outputs can be incredibly difficult because they are what economists call elastic. It means that they can change very easily. So you say, “I’m going to tax the energy for AI,” and maybe they shift the definition or skirt the definition of what AI means. Or you’re taxing specific inputs, and they maybe use different types of inputs instead. Can you elaborate on some of these problems with the actual tax base that we’re talking about?

Daniel Bunn: Yeah, absolutely. And there’s another element that I haven’t mentioned, which is where the innovation or the AI usage might happen. If the US adopted a tax on tokens and a tax on compute, and it was punitive to the extent that AI companies were like, “Well, we don’t need to be in the US. We can be elsewhere and everything’s accessible over the internet,” then that’s another margin for adjustment for these businesses. Now, they are putting a lot of equipment in the ground in the US, and that equipment is not super mobile, but it’s like saying at the early stages of the internet, “We know what a website should pay or owe,” without really knowing at the end of the day what the value proposition from the internet would be, or if there is any value at the end of the day. We need to run this experiment.

That’s a really important piece of this because a lot of these companies are running losses as they are building out their infrastructure to serve these new products and to serve the customers that are interested in them. That’s one of the things that we have as a feature of our tax system—it’s not my favorite part of our tax system, but we have a corporate tax on profits. What a lot of these policymakers and proposals want to do is short-circuit the idea that maybe we would tax profitable companies and instead say, “Hey, these companies aren’t getting money, but we’re going to figure out ways to tax them anyway and also hurt their ability to scale to any level of profitability.”

Adam Michel: The last piece of this debate that I want to touch on is this claim that we currently overtax labor in relation to capital, and that this disparity in the tax system will just make the impacts of AI that much worse. First, is this true, and how should we think about the relative tax treatment of capital and labor, or investment and work?

Daniel Bunn: Yeah, this is an area where, depending honestly on what side of the political aisle you come from, you’re going to see this very differently. The way I think about it is that with capital taxation, you’re almost always going to have multiple layers of tax on capital income. There might be entity-level taxes on businesses through corporate income taxes, and then you would also have dividends or capital gains taxes at the shareholder level. So if you want to compare the taxation of capital income versus labor income, you have to pay attention to all the layers on the capital income side. Now, some people, when they’re looking at the capital versus labor distinction, look at a different question than just the headline rates, and they say things like, “Well, because companies can deduct the cost of equipment, and if you’re buying servers, you’re buying a lot of equipment, then you’re essentially giving an advantage to the capital investment versus your workers.”

Well, the convenient thing for me is I can immediately say, “Well, you also deduct the cost of wages.” And where we have immediate expensing in current law in the US right now, it’s actually equal treatment on the timing as well. Now, the other piece of the equipment is you’re buying that equipment probably from another company who might be paying a tax on corporate profits on the sales of that equipment to you. There are other layers to this. But a lot of times, people who don’t appreciate the value of immediate expensing and wish the tax code did not have immediate expensing would say, “Well, the default should be that companies have to lose the value of their investment costs to inflation and to the time value of money, and that’s a better situation. Therefore, improving that situation is a bias towards capital investment.” I would just simply disagree with that.

Adam Michel: You laid it out well, right? So the business gets to deduct both labor and investment, generally speaking. And then on the input side, labor does face taxes, but then the return on capital also faces taxes. Those are sort of roughly equal if you calculate both layers of tax on the capital return: the capital gains and dividends tax, and the entity level corporate income tax. So I just think, at the end of the day, this is simply a way overstated case, and in most cases, it is not true that there’s any sort of systematic disadvantage one way or the other.

I want to move on from the tax side to the spending side. One of the things in the piece you wrote I thought was particularly interesting because with these proposals, the money often goes somewhere, into new programs. Can you just very quickly give us a little bit of history on trade adjustment assistance and how that may teach us something about mistakes we don’t want to repeat here?

Daniel Bunn: Yeah. So one of the things that policymakers love to do is they love to say that we’re going to find a way to offset this harm. Sometimes it’s well-intended, but it ends up being very difficult to identify the people who were actually harmed or to actually help them with the thing that you’re trying to help them with. Trade adjustment assistance is a program that’s now expired, but it was designed along with some of the trade agreement efforts in the US to say, “Hey, if we’re going to open our markets to another economy and agree to broader free trade with another country, there might be people who will be displaced by this open relationship.” This is something that economists know is possible and even likely—that some things will be more efficiently produced abroad than here. If we were producing those things here before the trade agreement and then after the trade agreement they’re produced abroad, then there might be some displaced workers, right? We know this, but policymakers see this and they’re like, “Well, we need to do something about this.”

This is a bit of the history of the trade adjustment assistance program where well-intended policymakers, who saw maybe a pot of money available for them to hand out to somebody who theoretically would be harmed by trade, ended up with very low uptake. The people who were harmed by this aren’t necessarily signing up for the program, and there are also challenging definitional issues. How difficult do you think it is, Adam, to identify that some worker in a certain city was harmed by trade versus that worker being harmed by the fact that the company just decided another city or state was the right place to be?

Adam Michel: And even harder with AI.

Daniel Bunn: Even harder with AI. Like, who’s going to lose their job to AI? Some people likely will, and maybe those jobs will disappear in different ways, but there will also be productivity enhancements and opportunities elsewhere in the economy. And so if you want to create some sort of sovereign wealth fund that’s supposed to fund maybe a universal basic income to everybody who’s been harmed or lost their job due to AI, I think you’re going to see similar things to what we saw with trade adjustment assistance. It’s really, I would say, a boondoggle waiting to happen, and it’s missing the point of what happens in a dynamic, growing economy. There are certainly people displaced, and certainly businesses grow and fail, and I don’t want to be insensitive to that. What I do want to do, though, is say that the policy solution is probably, at the very least, misguided.

Adam Michel: I think that on trade adjustment assistance, there’s also been long economic research on how it impacts people to go through the program. One of the things that you find is people that are eligible for the program, if they actually went through it, are worse off than people who didn’t go through it. That is largely just because you’re taking those people out of the labor market for a couple of years, resulting in lost wages, and they are not actively retraining in the dynamic economy. So often these sort of do-good programs that sound good on paper are causing more damage than if you simply had done nothing. And this is where—and Adam, I’m not sure where you stand on this—investments in our existing unemployment insurance program are probably a better use of policy innovation or policy solutions rather than saying we need this whole new program. There are existing pieces of, I would say broadly speaking, the safety net that are meant to provide a transition rather than a specific, “this person was affected by this circumstance” kind of approach that I think would be worthwhile looking at. Honestly, during COVID, there were a lot of people who were like, “Oh my goodness, the US unemployment insurance program needs some upgrades.” Certainly, let’s start by fixing the systems that we have that are designed for this rather than creating new things out of whole cloth.

Adam Michel: I think at the end of the day, you and I agree that AI will change the labor market, but what hasn’t changed or what won’t change is this relationship at the center of the whole debate that we’ve been talking about: that workers and the tools and the technology they work with ultimately make us more productive, and they are complements more often than not. Taxing those new inputs and taxing capital will most likely backfire. High taxes on AI won’t stop it; it will just shift it offshore or shift it somewhere else. The result will be less domestic investment and fewer American workers building and using the next generation of cutting-edge tools. So, Daniel, my final question to you from a sort of libertarian point of view—we are here at the Cato Institute after all—what is the sort of simple, quick takeaway from this conversation for you? If listeners remember sort of one thing about this AI tax debate, what would it be?

Daniel Bunn: Labor-saving technology is good. It’s going to change things, and change is sometimes scary, but labor-saving technology—and that’s what I see a lot in this AI debate—is a good thing. The other thing is that principled tax policy doesn’t get thrown out the window when something new shows up in the policy arena or in the economy. It is still the case that you should aim for simpler and neutral policies rather than saying, “Oh, well, you know, this specific thing, we’re going to tax it and ignore the rest of the economy.” Broadly neutral policy is the direction to go.

Adam Michel: My takeaway, in addition to those great points, is one at a higher level, and that’s: if you tax something, you get less of it. We want more innovation in the economy. We want more investment. We want people to be more productive and be able to earn higher wages. We want to see where this AI thing is going. I think it’s an optimistic story at the end of the day, and taxing it now is quite possibly the worst thing you could do. You’re going to stop that entrepreneurial innovation, and you’re going to stop the process that has made the United States such a great place to work, to live, and to raise a family. And so taxing it is just simply the worst way to respond to this type of economic change. Daniel Bunn, president of the Tax Foundation, thank you for joining me.

Daniel Bunn: Thank you.

Adam Michel: That wraps up today’s episode of the Cato Podcast. If you found this discussion interesting, please subscribe and share the show with anyone who might find it useful. You can explore more research, commentary, and upcoming events from the Cato Institute at Cato​.org. Thanks for listening.