The Case for a Public Share in AI

Jeremy Bearer-Friend: Our pitch is that these systemically important AI firms should pay a tax with stock. This would be in addition to whatever other taxes are still happening—payroll taxes, corporate income taxes, et cetera. But because of some of the unique features of this firm the public is entitled to a stake, to a share.

Rich Ford: This is Stanford Legal, where we look at the cases, questions, conflicts, and legal stories that affect us all every day. I’m Rich Ford. Please subscribe or follow this feed on your favorite podcast app. That way, you’ll have access to all of our new episodes as soon as they’re available.

I’m very happy today to welcome on to the show two of Stanford Law School’s esteemed alumni, Jeremy Bearer-Friend, a professor at George Washington Law School, and Sarah Polcz, a professor at UC Davis Law School.

They’ve known each other since back in their days at law school, and they continue to collaborate and now have recently published a paper in the Columbia Journal of Tax Law that outlines an innovative new way to address the potential harms and challenges posed by artificial intelligence. Also, to some extent, challenges posed by growing wealth inequality. And of course, we have heard that AI may contribute to and exacerbate problems of wealth inequality as time goes on. They plan to do this with a framework for a tax on AI companies, which would be paid in stock rather than in cash, and the stock placed in a public trust, so it would give the public a stake in AI.

Their research was recently cited by Senator Bernie Sanders in a New York Times op-ed and on a Senate floor speech. They’re now working with him to draft legislation creating what they call the AI Sovereign Wealth Fund, which would bring their proposal to life. And there are other senators and members of Congress who are interested in this innovative proposal as well.

Welcome, Jeremy and Sarah, to the show. 

Sarah Polcz: Thank you so much for having us. 

Jeremy Bearer-Friend: Great to be here. 

Rich Ford: Let’s start by taking a step back from the proposal and talk about how you came up with it. What motivated you to do this research? 

Sarah Polcz: The impact that AI is going to have on probably almost every sphere of human life has really become clear in the last few years. And not only in terms of will it create new wealth inequalities, new classes of billionaires, impact particular sectors where we might expect mass job loss. There’s been, we might say, theft, borrowing on a completely unprecedented scale of the human works, texts, and images that we’ve all created as part of our human culture.

And the value that’s been extracted from these was positioned so far to be captured, almost exclusively, by the tech sector. So we were really thinking what kind of approaches might allow us to tackle some of these differing questions in one particular policy proposal? 

Rich Ford: Yeah, so that’s…I work with a group called the Authors Guild, and we’ve been involved in some litigation against AI companies. You mentioned the theft of human work product, but sometimes pirated data has been used in order to train AI in order to then write and presumably take the jobs of authors, and this is something that I guess is quite widespread as an issue. So one answer has been litigation, but you’re proposing something quite different—a more proactive policy-based approach.

So you, Jeremy and Sarah, you were classmates at Stanford Law School and you knew each other and were friends. Have you collaborated before? And could you tell us a little bit about some of your early collaborations and how maybe they fed into this latest project? 

Jeremy Bearer-Friend: Yeah. I think our first project together, Sarah, you should correct me if I’m wrong, but was in a financial engineering class. There was a practitioner who was brought in with an expertise in financial crises, because that was the aftermath of the financial crisis, and we were going through all the different ways that financial engineering had damaged sectors of American life. And we, yeah, we collaborated, got honors in the class, I believe.

But a lot of it was even just shared interests and values. Sarah’s job talk at GW was on sharing in private ordering of artists—how do artists end up sharing the royalties from their work? So there’s already been this consistent theme around that in, in her work and from my standpoint, tax is how we use the law to share what we have. Tax is fundamentally about sharing. 

Rich Ford: That’s interesting. So there’s a set of IP issues that involves how you go about sharing the revenue generated from IP, perhaps among performers as opposed to songwriters, things like that. And that fed in, and then as well, the concerns about the financial crisis. We’re and who knows, we may be headed for another financial crisis of some kind provoked by AI and how you deal with a collapsed financial system in an equitable manner. So I guess that leads us pretty well into the details of the proposal. So tell us about the proposal.What exactly do you propose we do about the problems of AI?

Jeremy Bearer-Friend: So it’s really rather straightforward, and our pitch is that these systemically important AI firms should pay a tax with stock. This would be in addition to whatever other taxes are still happening—payroll taxes, corporate income taxes, et cetera. But because of some of the unique features of this firm the public is entitled to a stake, to a share.

And there are a number of appealing features to that type of tax, but one is simply it piggybacks on what the tech sector is already quite familiar with. These firms already divvy up ownership interests across their own employees, across founders and also with investors. So it’s actually relatively routine for them to do this. And it’s how many of the tech billionaires became billionaires. They were compensated with stock, and that’s where the real ballgame is, but the public sector has been left out of that. 

Rich Ford: Okay, so it’s a pretty…in one sense, it’s pretty straightforward because they’re used to giving stock options and stock-based compensations. This would just mean that the public sector gets a share, as well as the investors and the people working in the company, so it shouldn’t be complicated or hard to do.

Let’s talk a little bit more about why you think the public deserves a share in AI companies, because I’m sure you’ve heard some pushback from perhaps people working at AI companies about that idea.

Sarah Polcz: The interesting thing is less…well the bill hasn’t come out yet, but less pushback from AI as opposed to AI or tech sector as opposed to, say, the journalists at Wall Street Journal or the corners that you might expect. And what I think I…where I think education around how AI works can potentially have a role here, is that AI is very different from, say, the dot-com boom or the way that Google Search was created.

What AI has in common with other innovations is that, of course, it leverages prior knowledge and science, but we don’t normally think that entitles the public necessarily to an equity share. But the difference here is that while we use human knowledge for human engineering knowledge to design an airplane or to design the infrastructure for AI—what’s an airplane actually made out of? Metal, bolts, electronics components. What are the AI app models actually really made of? In a very real concrete sense, they’re made out of our words, our texts, reduced to tokens. They’re made out of real concrete human writings. I know that you’re an author. There are real works that can be pointed to that have all been ingested.

And what the tech companies have figured out how to capitalize on is the interactions between all of these words creates this surplus that is previously sort of unpriced in the market, you might say. And our position is that because really it’s not that you get the AI capabilities that we’re interested in until you’ve got this level of trillions of tokens, it really is a function of our collective cultural textual productivity that humanity deserves to benefit as well.

So of course, we would never argue that the developers of AI don’t deserve credit. Of course they do, and they’ll get plenty of it. They already have, and I think that they’re…what they’ve achieved is really inspiring and commendable. But the question is, should they capture all of the benefit, or should there actually be recognition of the fact that human society, human culture, images, texts are an absolute necessity for the production of this particular tool.

That’s been the lesson of the last 10 years of AI, is scale—data and scale. Not just more geniuses in the room, although that’s important. More data, more books, more articles, more images. And that’s been really what has led to these just really extremely impressive abilities of the modern models. 

Jeremy Bearer-Friend: But, and because of the…

Rich Ford: So there’s something right now on, just from an IP perspective, right now the position of the AI companies, as I understand it, is that all of this is fair use. And so there’s no requirement of compensation, and the authors can’t refuse to grant access because the AI models are using so little of the author’s work that it’s just fair use. Provided, of course, they buy the books. They can’t use pirated books, but they can just buy the books, plug them into AI, and that’s fair use.

And it seems like your position is that this is different than my reading someone else’s book, being inspired by it, and then creating my own work. That’s not exactly what AI is doing. AI is doing something that’s, kind of, because of the scale, and perhaps also because of the way the technology works, something that’s fundamentally different than that, and that our current IP system just doesn’t have a way to deal with. 

Sarah Polcz: That’s right. So we’re not at all suggesting that the equity share would compensate, you know, authors exclusively, or that it should take the place of litigation, which is, of course, it’s the position of the AI companies that it’s fair use. Others would dispute that. But that can all play out, and authors who can find that their works have been trained on can pursue those claims. And we’re not proposing that this take the place of that…of that process at all. We’re talking here about this surplus value that comes from the scale, that comes from trillions of tokens.

It was previously unpriced because the technology and techniques didn’t exist for extracting value from it. But given that those have been created by the founders and the AI scientists and their…through their tremendous work, given that value can now be extracted, the public, as providers of the core input to that, deserve to have a share of the benefits that are derived from it.

Jeremy Bearer-Friend: And tax is a natural tool for redistributing that, right? Tax is how you would share that across a whole society. 

Rich Ford: Great. So yeah, let’s talk a little bit more about the tax side of things. So what exactly, do you have a proposal of exactly how large a share the public would get? And how does this compare to other types of tax proposals? Because I…correct me if I’m wrong, but typically a tax proposal is going to be a percentage of earnings, a percentage of income, something along that line, and this is actually an equity share in the enterprise. Is that unique to this proposal or are there other tax proposals that have done something similar in the past?

Jeremy Bearer-Friend: Well, if we look to some of the earliest traces of tax in human record, right? If you were to go to Torah, for example, my parsha portion at my bar mitzvah ended up being tax related. It was that the corners of your field needed to be given to the poor and to immigrants, and so you shouldn’t harvest that side of the field.

So the very initial concept of a percentage of what you have needs to be shared with others wasn’t about cash, right? It was about the yield from this crop. So I don’t think conceptually we’ve ever really been in this cash box. But in a lot of the policy debate, it’s just been this tacit presumption that the payment should be assessed and remitted in cash. But that’s an unnecessary assumption.

The alternative to what we propose of this tax paid in stock is a currently defective income tax system. So we’ve seen that both the billionaires who are profiting off of this are, have lower effective rates than most members of their own company who are working for it.

And then we see that the companies themselves are typically loss firms because they are spending so much on R&D they really aren’t going to have any earnings until they’re already monopolies. This already was the case with Amazon, this was the case with Uber, this was the case with Meta. So our corporate income tax is just not doing the work of taxing them, and our individual income tax is not doing the work, so this helps supplement that. But it’s not meant to replace any other tax. We can have a diverse tax system, but it is a nice backstop because it’s obvious that there’s value here. If we were to assess this tax, the principal economic incidence is on the shareholders, and the shareholders are going to see diluted value in what they hold, and depending on the rate, that could be quite substantial.

Rich Ford: I see. So do you have a sense of what the rate should be, or is that for further research to determine? 

Jeremy Bearer-Friend: You know, in our article we were agnostic about it. We say a “non-controlling interest” in our published work, but once you’ve designed a tax mechanism, then the rate is just a dial, and it can go where you want it to.

In what Senator Bernie Sanders has proposed, he puts forth a rate of 50% which is not in our writing. It’s gotten a lot of traction, I will say, that has been quite appealing to a number of audiences. But it raises other questions, especially around the governance of the fund. If the interest is so concentrated, then governance questions matter a bit more. 

Rich Ford: Right, of course. So if the government has close to a controlling share or a controlling share, then they’re going to have governance rights that they wouldn’t have if the share were 10% or 15%. And that raises questions about the government getting control of AI as opposed to just getting part of the revenue.

So you’ve mentioned Senator Sanders. What has the reaction been among politicians, and also what’s the reaction been among AI executives? In one sense, this sounds a little bit like what Sam Altman of OpenAI proposed at one point relatively recently. That, you know, he said AI’s going to take a lot of jobs, and we’re going to need to find a way to protect people whose jobs have been displaced, and the public ought to have a share. So, maybe the industry is…might be behind this? 

Sarah Polcz: One thing that I would say on that point is that I’m not surprised that the idea of the equity share would resonate with Sam Altman because he’s someone who, of course, understands how it’s built, right? As I was talking about the contribution that all of our texts, images, novels, articles—he understands just how central that is to the capabilities of AI. And given that core contribution, more of an ownership interest as opposed to UBI or a handout or just corporate philanthropy, it’s less appropriate. The standing of the public should really be more on the level of, equally in the category of an owner. But Jeremy, maybe you want to speak about the structure of his proposal compared to that of Senator Sanders.

Jeremy Bearer-Friend: Well, the most recent reporting was just from last week when Trump and Altman met together and tried to co-opt the proposal. I think it was within 48 hours of when Bernie had announced. But the proposal that they made public was worrisome for two reasons. One, a tax is obligatory, and what they propose is voluntary which means that it’s fundamentally a quid pro quo arrangement.

They are planning to just do private deal-making firm to firm to firm, and then it just feeds into the primary dynamic of this administration, which is self-enrichment. I think it would be a spigot of corruption. So that is not what Bernie is proposing there. The fund would go to the public, and there would be transparency, there would be conflict of interest rules, some of the accountability we would expect of a public institution.

Rich Ford: So that’s interesting that there, to some extent the Trump administration has kind of accepted a premise of the proposal, but they’ve tried to turn it into a series of one-off deals as opposed to a general tax obligation. Any reactions from other people on both sides of the aisle, other than Senator Sanders, about the proposal?

Jeremy Bearer-Friend: Well, I would just say quickly one other distinction from Trump is also he’s pumping public dollars into these firms because with his deal with Intel, for example, he actually is handing over public money to a company that’s already very heavily subsidized through our tax code. So there’s been a lot of tax expenditures for them and yet in, through a Trump arrangement, it’s actually giving them cash as opposed to a tax which has to be enacted by Congress. Congress has the taxing power and a tax which wouldn’t require giving public money over to this sector. 

Rich Ford: This is Stanford Legal. I’m Rich Ford, and we’re here with Jeremy Bearer-Friend and Sarah Polcz talking about a proposal to give the public a stake in AI. Jeremy and Sarah, you mentioned that there was some pushback which I’m not surprised to hear. Could you tell us a little bit about the objections to this proposal and how you’d respond to them? 

Sarah Polcz: Sure. One, we touched on more briefly before, which was, what’s different about AI? All of our technologies, most new initiatives are building on past knowledge, past scientific contributions. We usually just think that’s free to use. And so I was emphasizing that, AI’s different in the way that it very concretely uses human contributions, where you can point to the exact text tokens that have been included in there. And there, it… I feel like it becomes more compelling the more the technology improves.

The more human-like it becomes, the more you can really see how it’s a product of everything about us that it’s taken in. So I feel like that just becomes more… I feel like that critique diminishes in impact as the technology improves. 

Jeremy Bearer-Friend: Just another aspect of the pushback that’s actually been kind of thrilling to see is, at this point, I think within the first week of Bernie announcing he wanted to move forward with the idea, Washington Post did an op-ed about it, Wall Street Journal did an op-ed about it, Financial Times did an op-ed about it, Fox Business News did a segment on it. Glenn Beck did a podcast about it. It just made so many ripples, and I think it’s because they can see, even the critics can see the viability and the possibility. So…but the piece that rings most true to me, I would say, is the political insulation. In our current moment, I am concerned with fascism. I don’t want to see a consolidation of more power under this specific administration. And so it would be important in legislation that there be some firewalls against that direct control from the executive. 

Rich Ford: Okay. So one of the main objections might be this is public ownership of a major industry, and that gives rise to all of the concerns about consolidation of power, particularly in the hands of a government that could turn authoritarian.

So…but your proposal is, you know, although of course the government has to administer this, this would be held in trust for the people. So let’s talk about the difference between something where it’s government control and held in trust for the people, and how do we make sure that the money is in fact administered for the benefit of the people as a whole, and not for individuals or particular administrations?

Jeremy Bearer-Friend: And one of the best examples of that is over in California. So the CalPERS fund has over $300 billion of assets under management, and they have a duty to then…to protect the interests of beneficiaries, which are public workers, public sector workers in the state of California. So in that environment, you could worry, “Oh no now California isn’t going to regulate companies anymore because it’s worried about the profits of its retirement fund,” but that has not been the case. The legislature has been happy to regulate industry in California, and the executive branch as well. So there is this path that has already been pretty well-developed, and I think we can draw from that model for this. 

Sarah Polcz: I would add one more point to that which Senator Sanders raised, which is that we can try and put in place protections and at least under this approach there’s some democratic process. Whereas really, we should be comparing it against the status quo, which is that the roll out of the technology and decisions about whether or not its benefits are to be distributed broadly across the public, more broadly than would be the case or is the case under the current system. Those decisions currently rest in the hands of the executives at the AI companies.

Rich Ford: Right. So we’ve got a model in place that we could use in order to insulate this from overt political manipulation and at the same time, the alternative is to leave it in the hands of a handful of tech executives given that everyone understands that this technology is going to be in some way quite transformative. I wonder, given the…in a sense, this proposal is extremely simple. It’s a very elegant solution to what looks like a very large and almost intractable set of problems and novel set of problems, and that’s a great strength. I wonder, given that, whether you have ideas for additional applications to this kind of approach to dealing with the new technologies and with the growing disparities of wealth. 

Jeremy Bearer-Friend: Well, I would love to see some bipartisan versions of this legislation. I think Senator Sanders is absolutely getting the idea out there and making it well-known, but there’s still potential for other offices, perhaps in the House to work across party lines given the popularity.

I also think, and just speaking for myself, that there’s a real opportunity at the state level. California has jurisdiction over these AI firms. It has a more functional legislature, and there’s a real demand to tax AI in the state. So an appeal of doing it at the California, at the state level, is also you have federalism as a bit of a firewall against fascism that there could be state control, and that’s not going to be up to the current presidential administration.

Rich Ford: So it’s possible that we could start at the state level, but also you consider the state level to perhaps be a bulwark against possibly problematic federal government or federal administration.

Well, thanks so much. Any last thoughts or comments, anything that I missed that you think our audience needs to hear about this proposal?

Sarah Polcz: Just stay tuned I suppose, we’ve already been pleasantly surprised by the support from the president, the support from Senator Sanders, and hearing from other representatives. So it’ll be interesting to see how it unfolds alongside other proposals. 

Rich Ford: Well, Jeremy Bearer-Friend and Sarah Polcz, thank you so much for being on the show. 

Jeremy Bearer-Friend: Thank you. 

Sarah Polcz: Thanks for having us. 

Rich Ford: Thanks to Jeremy Bearer-Friend and Sarah Polcz.

This is Stanford Legal. And if you’re enjoying the show, please tell a friend and leave us a rating or review on your favorite podcast app. Your feedback improves the show and helps new listeners to discover us. I’m Rich Ford, see you next time.