Would Public Ownership Push Us Towards AI Sovereignty? And Will We Ever Get Sensible AI Regulation?

Todd Tucker, my long-time sparring partner on international economic policy issues, has a new piece with his colleague Elizabeth Wilkins called "How Public Equity Can Promote a Revived American Democracy." It's behind a paywall so I couldn't read it, but when he explained it in a Bluesky thread, he brought up public equity in the context of the AI sector:

The latest idea in this space is @BernieSanders' proposal to deposit 50% of the shares of leading AI companies into a new sovereign wealth fund.

While it could mean $$, the bigger reason to consider this is greater democratic control over our era's most disruptive technology.

I wasn't looking to engage on the merits of public equity in general (I'm against it!), but the international aspects of the AI part of his argument intrigued me, so I asked him whether his argument for public equity here applied only to the U.S. government or to other governments as well:

With regard to AI companies in particular, is your argument that the US government should take equity stakes in domestic AI companies, but no other government should do this? Or do you think other governments should also adopt this policy and take equity stakes in domestic AI companies?

His replied as follows:

Any country could certainly demand this of any AI company as a condition for operating domestically.

But a sustainable policy regime internationally (which we do not even attempt to get into) would probably involve some type of exclusivity or "apportionment" where home countries get more of a claim.

I followed up with a question about whether his suggestion might lead to some economic nationalism for people choosing between AI companies:

Do you think such a policy regime is likely to lead to some degree of national boundaries for AI, under which citizens of a country use the services of national AI companies because they (1) share in the profit from these companies & (2) have more trust in the way these companies handle their data?

And he answered:

Possibly. But I think we're already seeing some degree of that desire, even without the policy.

Todd and I disagree on a lot in terms of what economic policy should look like, but sometimes we agree on what things do look like, and here I think he's right about what we are seeing at the moment. There is a growing awareness of the ties between the leading tech companies and the governments of the countries where they are based, including how governments can access consumer data and how they can exert control over who uses the products/services. It seems to me that an absence of trust in these governments is likely to intensify the push for homegrown alternatives. (In the U.S., the main concern is China. In other parts of the world, there is growing concern about the U.S. as well.)

Tech sovereignty vs. concerns with specific governments

This is playing out in the policy debate as "tech sovereignty," with some companies pitching their services as having a national orientation, which they hope will make them more trustworthy for the citizens of the nation(s) where they are based. I've seen this in the social media world with both Canadian and EU companies taking this approach.

But the situation goes beyond pure nationalism. In addition to a call to use domestic providers of services, there can be a broader pitch to consumers that is more directly focused on privacy and other guarantees, as well as an assurance that use won't be cut off by the government where the company is located. Many consumers will prefer to use products/services for which they can trust the government that is setting the rules. Having data and operations hosted in particular countries can therefore be a selling point to consumers around the world.

While I think Todd is right that all of this is happening already, I would say that a shift to public ownership could further stimulate this trend. For example, Europeans who are currently worried about using OpenAI's ChatGPT – e.g., for privacy reasons or a fear of becoming dependent on ChatGPT and the U.S. government then cutting them off – would likely be even more concerned if the U.S. government owned shares in OpenAI.

These issues will have an impact on the international competition among AI companies, a key theme of which today is the U.S.-China competition to "win" the AI race. While it will be interesting to see which country's industry will have the most advanced AI, I think that is only part of the race that's happening here. Currently, U.S. AI companies have the most capable models, but it seems likely that eventually we will get to a point where the basic models can do everything most people want or need, and the advantage held by the leading models will be less noticeable and valuable. Anthropic's Claude may be the best, but for the average person second or third best will be more than enough. At that point, the choice of model for many people is likely to be based on factors such as trust in the company (and cost, of course), with trust linked to the country in which the company is located and the policies/regulations that are in place there. And public ownership of the company would likely play a role in the level of trust as well.

Public ownership vs. regulation of the AI sector

As to Todd's main point, I'm not sure public ownership necessarily means "greater democratic control." Certainly it means strong bureaucratic oversight, but I don't think that's the same thing. A small number of experts working in government who make decisions on complicated technical issues without much transparency is not necessarily the democratic ideal. A key question here is, greater democratic control as compared to what? Would public ownership be better than detailed and comprehensive legislation and regulation of private companies?

To me, what is needed here is a sensible version of the latter. But the companies don't want that, and I'm not sure legislators and regulators are able to provide it at the moment. At some point they will figure it out, but it may take a crisis or two to get us there. In the meantime, there is plenty of room for experimentation among governments around the world, and hopefully some of them can move things forward a bit.

As is probably obvious to this blog's readers from 20 years of me posting, in many areas I'm a skeptic on regulation. But here the potential harms are wide ranging, and we are likely to need something pretty extensive. AI regulation will have to address a much broader set of concerns than, say, automobile regulations do (of course, we now have to deal with AI in automobiles!). From what I can tell, the industry doesn't seem like it has a handle on this, so it's going to be up to legislators and regulators. Perhaps the courts can help too (a suggestion I recently came across for applying strict liability here was interesting), and there is plenty of litigation going on, but it would be nice to have a coherent policy rather than a hodgepodge arising from court rulings.

My own experience with AI is still pretty limited, so what do I know?!

As for me personally, since I last posted about this, I've actually used AI tools for a couple tasks: Trying to figure out some problems with our website code and coming up with a market value for a house. It seemed moderately useful for those tasks, although with the website problems it took a developer looking deep inside the code to get things working again. (I should note, though, that at one point the developer was using AI to help diagnose the problem, so I shouldn't totally dismiss AI in that case).

With regard to law/policy thinking and writing, I've heard people say they use AI for research or brainstorming or drafting. I'm never going to use it for any of those tasks because they are things I enjoy doing the way I do them now. Perhaps AI will someday be able to eat pepperoni pizza for me, but that's an experience I want to keep enjoying and I'm never going to let AI eat my pizza! I feel the same way about my current process for thinking and writing.

But I can see how someone might not always want to do one or more aspects of these tasks, and how outsourcing that to AI might make sense. For that reason, I'm not an AI-denialist. I see the value, although I think it is perhaps overstated. In addition, I suspect that when people have to pay the actual costs of AI tools instead of getting them at the current subsidized price being offered by the AI companies, things might shift a bit. And there are big copyright issues still to be dealt with here, which could change what AI companies can offer. For these and other reasons, I'm happy to stay mostly on the sidelines for now and let the companies and governments work out what AI will turn out to be.