Dean Ball (6:18)

Thanks so much for having me.

Derek Thompson (6:21)

Tell me a little bit about your time with the Trump administration. When did you join, what did you do there, and why did you leave?

Dean Ball (6:28)

So I joined in April of 2025, and basically I was a senior policy advisor for artificial intelligence and emerging technology at the White House Office of Science and Technology Policy. So those are very, White House staff roles are very capacious and sort of day to day changes radically depending on what's going on. And so I did, you know, quite a bit, quite a bit of stuff. But the main thrust of my work was to play a role in the drafting of the administration's AI action plan and some of the other policy moves that were attached to the action plan, such as the executive orders that the President signed when he, when he announced the action plan. So those were the things I worked on. I worked on them, of course, with many other people, you know, so a lot of what I did was like shepherding these documents through the interagency approval processes that these kinds of things must go through. And then I left in August of 2025. And principally I left because, you know, I, I feel like my primary value add is as an independent sort of writer, you know, sort of thinker. And, and the action plan had been, you know, completed at that point. It was sort of rolling downhill in the sense that the bureaucracy liked it, the bureaucracy was excited about adopting, you know, about, about implementing it. And so I didn't feel like this. I felt like the sales and communications job was largely done. And I said, okay, well, you know, my value add is probably doing other things now.

Derek Thompson (8:03)

Do you give people a sense of the, the substance of the action plan and the substance of the executive orders? Maybe talk about it this way. What are the most important differences between the way that the Biden administration and the Trump administration thought about artificial intelligence?

Dean Ball (8:19)

The Biden administration, I would say, took a far, you know, a far more, I would say an ominous approach from a regulatory perspective. Right? There was like, way more foreshadowing of regulation to come. I think it's exaggerated sometimes how much they themselves actually regulated. But if you look, if you go back and look at what they did, they were clearly putting the scaffolding in place for a significant regulatory regime to come. Probably, you know, in, in the, the term that, that Joe Biden, Joe Biden's second term or Kamala Harris is first right so the Trump administration, first of all dismantled large portions of that. Second of all, I think thematically the big difference is, you know, there's sort of a vibe based difference of, you know, we want to this technology, we want to let it grow, innovation, et cetera. But one other one that I think is really important is this issue of adoption and diffusion. So I think the action plan, if you look at it, one of the things you'll see that's sort of laced throughout it is this emphasis on transformative adoption of AI throughout the economy and in government. And I think the Biden folks were, on the other hand, much more interested in. This idea of very, very large models trained on huge amounts of compute in these giant data centers and like the development of those things and where those data centers would be and you know, they'd be on federal lands with big military installations. And you know, again, like Trump administration wants to build those data centers and like train the big models and all that stuff. It's not about that. It's more about like this idea that what we're doing is we're racing toward this development of some big model after which everything is different. So way more of a Manhattan Project kind of vibe. Whereas I think the folks in the Trump administration certainly myself kind of don't see it that way. Sort of see it much more as like a. This is about the challenging issue of diffusing this technology and transforming the way that work is done, the way that organizations are structured, et cetera, et cetera.

Derek Thompson (10:39)

One difference that I've observed is that Trump's economic centerpiece is his tariff policy. And I don't think people appreciate just how many carve outs and exemptions there are for artificial intelligence and the computer parts that go into the building of artificial intelligence in that tariff policy, it is billions and billions of dollars, tens of billions of dollars of carve outs. So AI isn't just exempt from the tariffs. We are also, and you mentioned this at the end of your last answer, actively promoting the sale of AI chips in other countries, including in some cases China, in a way that the Biden administration wasn't. Because to your point, there was a little bit more forward looking fear about what would happen if we sold these chips all over the world. In a strange way, is it too cheeky of me to suggest that Trump's AI policy is slightly more globalist and neoliberal than the Biden administration's given how much they care about the diffusion of the sale of AI chips all over the world, including into China, which is something the Biden administration was specifically worried about and trying to block.

Dean Ball (11:50)

Yeah, I mean, I think the way that, the way that they would put it, the way that people in the Trump administration would put it, would be that they come at this from the perspective of Silicon Valley, and they come at this from the perspective of we need to build, build global ecosystems of, you know, around our chip technology, right. In the case of the export controls, things like Nvidia's CUDA development environment for AI applications and AI models and whatnot. And that, like these, these types of things are, are really important parts of this is how you establish tech dominance in the long run. Right? I think that's kind of the idea. And so, you know, in that sense, though, like you said, I said global ecosystems, you said globalists. Tomato, tomato a little bit, right.

Derek Thompson (12:41)

This is why I wanted to start here before we get into the showdown between Anthropic and the Department of War, because it presents, I think, a really ironic grounding for this showdown. You have this Trump administration come in that is more business friendly, that is more capitalist, that is more interested in the globalization of this technology, more of like a direct pipeline of Silicon Valley Straight to D.C. i mean, David Sacks, who's in charge fully of AI policy in the White House, is, you know, a co host of the podcast of Silicon Valley. And that brings us to the doorstep of this showdown between Anthropic and the Department of War, which I described in the open. And you know all too well, why do you think this showdown and the announcement of supply chain restrictions, why do you think this was such a remarkable, an important move from the White House?

Dean Ball (13:36)

So, for, for a lot of different reasons, frankly. But I think the sort of, the most important thing that I see here is this is a, this is a whole different level of. This is not just a contractual dispute. This is the government saying, if we can't resolve a contractual dispute, we are going to force you to do terms on our business by essentially threatening, you're existentially threatening your company. This is a general purpose type of threat that can be, you know, there's nothing in principle that stops this from being extended to other, to other types of businesses or other people in the AI industry. And so it just sends, it creates a tremendous chilling effect. And I also think that, you know, if you want to talk about like the AI race or the competition between the United States and China, one of the reasons that it's very hard to do business with Chinese companies, both for American firms, but also globally, is that everybody knows that Chinese companies are considered by the government to be military assets. And so if you use Deep Seek and the Chinese military wants to know what all the Americans are, you know, talking to Deep Seek about, talking to Deep Seek's models about, they will just. Well, it is not a question of can they get that data that is like 100%. Yes, they can. And everyone in the world knows it. This is common knowledge. This is not common knowledge about American companies. If the government demanded that Google turn over all Google search data, Google would say no. They would absolutely say no to that. This is eroding that level of trust. This allows because of the independence of American businesses, not total, but significantly greater independence of American businesses from the US Military. It's easier to trust and work with them. We were already having trust problems with American AI, this whole notion of AI sovereignty and whatnot. And I think now those trust issues are, are considerably worse because our government is behaving with the same attitude that China's behaves with.

Derek Thompson (16:05)

I mean, you said in a great essay, Claude C L A W E D on substack, that this was akin to an almost Chinese Maoist move to remove private property rights on behalf of a company. I mean, it was. It's that significant. It's that clear an attack on freedom.

Dean Ball (16:26)

Yeah, I mean, I think in the limit, right? There's no, there's nothing in principle that stops the government from saying, you know, you have to agree to our contract terms. And I see people from, from, you know, defending this move, saying, well, of course, like, you know, you can't set. The businesses can't set the terms of their engagement with the government. If the government, the military needs it for national security, of course they have to provide it. And you know, one of the terms of a contract is price, right? So like, if you're saying that the government, businesses can't set the terms, like you're saying something quite radical. And it's like, the thing is, I know people who are saying that don't believe it. And they just haven't thought about this, you know, in a principled way. And part of the reason for that, I fear, you know, America was founded kind of uniquely. America is a country that was founded on principled thinking about, like, you know, okay, how should this work in a timeless and universal way? Right. How do we create rules, procedural rules that, that, that can withstand the test of time and that will like, always be right. And I fear that a lot of the people that are defending this kind of just either explicitly don't Understand this about America or they don't care. They think that that's long since dead, which was kind of what the Claude essay was about. They kind of just think that that's, that's gone and like, we don't do that anymore. And now we live in the jungle, the dog eat dog jungle. And like, if the government wants your stuff, well, maybe you can fight them, maybe you can't. But yeah, I mean, that's, that's, that's kind of the terrifying aspect of all this to me. And so I don't think the government is saying here, like, I don't think they're truly doing a Maoist seizure of all the private property in this moment. But what it suggests, and as an American, I think you do have to think about principles and procedures. What it suggests is not positive for the trajectory of private property rights in this country.

Derek Thompson (18:20)

I've spoken to folks from the Trump administration and the case that they would make. The reason that they think they're on solid ground with this attempt to essentially nuke Anthropic from orbit is that they'll say, look, if Lockheed Martin sells the US Government a fighter jet, they do not have the right to say, and oh, by the way, don't use this jet to bomb Iran.

Dean Ball (18:40)

Right.

Derek Thompson (18:41)

Don't fly in the Middle East. Anthropic does not have the right to request, case by case, permission to use a technology that is in the employ of our military national security systems. And to take the dramatic example that I heard over and over again, this is what I kept hearing. If China fires a hypersonic missile at the US and we need AI directed autonomous technology that uses Claude Anthropic to disrupt that missile, we cannot call, or have to call Dario Amadei's office and get put on hold while this hypersonic missile is speeding toward an aircraft in the Pacific to get permission from Anthropic to use their technology to defend America. We're currently in the very, very strange situation where I am representing the Trump administration, arguing against a former member of the Trump administration, but continuing this play act. How do you respond to, to that defense of the Pentagon?

Dean Ball (19:36)

Well, so I don't think it's crazy for, for the, I don't think the principle that the Pentagon is standing on here in broad strokes is, is, is crazy. I think it's, it is a little bit. One of the jokes I make about new tech, right, is like they're kind of new to thinking about any matter of civics, like thinking about this at all, because it's like, it's like, well, first of all, like, defense contracts have plenty of usage restrictions in them. Like, I guarantee you that the, the defense contract of the, the contract, you know, between the Department of Defense and, you know, Lockheed Martin for the F35 or something like, does not say, you can do whatever you want with this, sir. Thank you so much. I'm so scared of you. You're so strong. Right? Like, that's not what it says. I think what it probably says is like, there's probably, like, it's probably extensive and has like an extensive list of like, intellectual property restrictions and usage restrictions. Things like, you know, hey, if you fly this thing above X, Y, Z altitude, it's going to break and we can't be responsible for fixing it. Right. We, we make no representations that it can exceed these speeds or exceed these G forces or whatever else. Right. And I'm sure that is extensively lawyered and detailed in the contract. But at the same time, like, I think it's not unfair for the administration to say, yeah, that's fine, but like Lockheed saying that that's like a purely technical restriction that has to do with the, like, physical characteristics of the jet that's quite different from, you know, some of these anthropic restrictions, which feel more like public policy types of restrictions. I still don't know how abnormal restrictions like that are. One of the things that I think gets lost in this debate is that the Trump administration itself, like the Trump Department of Defense, before it was called the Department of War in July of 2025, agreed to a contract with these restrictions in it. So it's not like, oh, this is some incredibly beyond the pale thing. Like, sort of seems like, you know, a lot of people looked at this, including like, you know, DOD lawyers and stuff. Right. Like, there's a lot of people with a lot of experience in looking at contracts who weren't like, oh, this is super abnormal. Right. So it does feel a little bit like we're, we're freaking out about something and it's, it's perfectly fine to freak out about something that, you know, previously people didn't freak out about, but I think we need that, that context is valuable to have. The last thing I would say specifically on the, like, Dario has to. The Pentagon has to call Dario. You know, I've seen that reported that Daario explicitly asked for that. I don't think that's true. I'm just going to come out and say that. I just simply think that there's either context missing here. Look, a Lot of people are making. I've heard a lot of people make claims about what both sides have said in this that can't possibly both be true. So, like, someone is lying here, and I'm not going to make a general claim as to who is lying, but I do think that, like, I would just be shocked if the actual thing that, like, the CEO of Anthropic said was, I want you to call me in the event that there are hypersonic missiles. It just. That doesn't sound. It's like one of those. There's jokes that people make sometimes about columnists where, like, a guy like David Brooks will just like. Or Tom Friedman famously will, like, make up interactions that they had with, like, cab drivers or whatever. So, so as to illustrate a point. And this feels like that. It just feels. It feels. It doesn't feel like an interaction that real human beings had just stepping out

Derek Thompson (23:15)

of the play act here. I have the same feeling that the Gotta call Dario Amade's executive assistant and get put on hold when the hypersonic missile is passing toward the US Felt a little bit to me like a hypothetical that had slowly gained the weight of fact, if that makes sense. That, like, it started as a hypothetical in terms of arguing for the Department of War's position, and then as it kept getting repeated, it was like, oh, wait, is this.

Dean Ball (23:40)

Did.

Derek Thompson (23:40)

Did this actually happen? Was this actually, like, a point of contention? Or are we talking a little in the abstract here?

Dean Ball (23:47)

My understanding also, by the way, is that just as a very quick point of fact, my understanding is that Anthropic offered, and this is based on public reporting that they offered for there to be a, you know, an exception to the autonomous lethal weapons clause for autonomous air defense. Though it's also worth noting that autonomous lethal weapons, like the kind of. The kind of air defense equipment that you would use to. To destroy incoming hypersonic missiles, wouldn't count because they don't kill people, they kill missiles.

Derek Thompson (24:19)

So. Right. Another. Another defense in the administration. And I just, I think it's useful to test your position here against some of the more popular defenses. Ben Thompson, the famous tech columnist, author of Shore, I think someone we're both a huge fan of.

Dean Ball (24:33)

Yes.

Derek Thompson (24:33)

Has said, look, the AI Frontier labs have gone around claiming that their technology is akin to the nuclear weapon of the future, a digital atomic bomb. You cannot possibly be surprised if you go around saying you're building atomic bomb for the government to want extraordinary amounts of control over your technology. After all, the Manhattan Project was not some project. We farmed out to GM and Ford and said, can you please both work on the project to build an atomic weapon. And once you've refined enough uranium and have blasted this thing in the middle of the desert, America let us know and then we'll buy the bomb from you. No, we did not do that. We had the Manhattan Project. We created an ultra top secret, highly secured laboratory environment in New Mexico, developed the bomb there. It was all under the thumb of the US military. If this is the 21st century version of that, then these labs can't possibly be surprised when extraordinary measures are used to control their dangerous, or soon to be dangerous technology. How do you feel about the Ben Thompson defense of the Department of War here?

Dean Ball (25:49)

Well, I find it to be. Truthfully, I think it's not entirely crazy. But I also, there's a part of me that finds it slightly disingenuous. And I'll tell you why, because I don't know if Ben has ever said this explicitly, but a lot of people who have said this, who have made this type of argument are people that would have also spent the last two or three years saying that anthropic is lobbying for regulatory capture by supporting the idea of technocratic regulation on frontier AI. And so you kind of can't have it both ways. It's like, what did anthropic expect would happen? It's like, well, maybe they expected that there would be technocratic regulation on frontier AI Maybe exactly what they've been saying, like, hey, there are national security implications here. We should have regulation about that. And like we think we're not doing, you know, nearly enough on the governance front. And we, we need the government to be way more engaged. Right? Like that is what they've been saying. And that doesn't mean they're saying please, you know, force us to do. They're not saying please seize our property. Right? They're saying please regulate us. But when they say please regulate us, it's accused of regulatory capture. And then when the government says we're going to seize your property, it's like, well, what did you expect? And it's like, yeah, I just think that it's attributing maximal bad faith to them on sort of both sides. And I think that's just not consonant with the facts. The other problem with the nuclear analogy that makes this all so difficult is that like these, these technologies are going to be so useful to the everyday American, to all of us, that it feels like, and also like I just feel as though, I just feel it in my bones. And I think everyone does at this point. How can you not look at this on. Look at this technology honestly, and say to yourself, you know that this will not be an incredibly important part of. About how all of us exercise our liberty in the coming decades?

Derek Thompson (28:00)

Let me stop you right there. I think there's actually huge disagreement on that point. Just a chasm of disagreement. And, in fact, I think you have pointed out before that there's a huge gap between people who feel like artificial intelligence is maybe the most important technology ever developed by humankind and people who basically look at this and say, I think this is somewhere between vaporware, a scam, and a technology that is incredibly facile at creating AI slop that is filling up my Instagram feed. There are millions of those people. I know for sure that there's a lot of them, because I see the comments and the emails in every AI podcast that I do. And that's where I do think that the policy fight that we've been talking about is really part of a much bigger philosophical disagreement. One part of that disagreement. Let me just set this up for you, because I know that this is a home game for you. This question that's coming down the pike. One question at stake here is, is AI a normal technology? You believe that we live in a world where AI is about to have a takeoff moment that makes it unlike any technology in modern history. The Trump administration, it seems to me, is acting like an administration, a group of people that doesn't think we live in that world. They think this is a normal technology. They think this is a computer. Tell me a little bit. Tell me a little bit about why you think this technology is abnormal.

Dean Ball (29:33)

Well, there's so much to say here. First of all, I think that AI as a normal technology is an essay by two Princeton University scholars, and its headline became a meme. A meme that I think has justified a lot of very poor strategic thought with respect to artificial intelligence. Their point was not. AI is not an utterly transformative technology that will change the architecture of the world. Their point was more. It was supposed to be a call to action. It was supposed to say, we have agency to shape the trajectory of how this goes, but we have to do that. We have to actually do it. Traffic law didn't invent itself, right? Stoplights didn't invent themselves. Someone had to go and come up with the idea and wire them in and pass the laws that you have to follow. Traffic laws, all that stuff that took a lot of effort, and that didn't just happen. And it also wasn't entirely just the private sector that did that. So this notion that macro inventions have these profound public civic implications and therefore the whole of society has to respond, I think that was what. AI is normal to technologies about it. In that sense, I concur with AI as a normal technology. I do not, however, concur with the meme of, you know, this is like a new form of the relational database or this is like a new app store. Like. No, and I. The thing is, is that I think when you're actually in the trenches, the governor. I think the. I think it's very easy to say that when you're trying to justify various policy outcomes. But I also think that like, when you're in the trenches and you're the one, like especially in the military context, the military, the, the, the, the, the. The DOD is not behaving as though this is a normal technology. Right. Or Department of War is not. Right. Now they might, well, they might say that, right? Rhetorically. They might say, yeah, you know, we don't believe in AGI or whatever, but like what. They're part of the claim that they're making. And this is something that, you know, your, your co author Ezra Klein has pointed out to me, is that like, so I'm actually taking a point from him here, but like, one of the claims being made is the reason we need to do this, like supply chain risk designation that is existential for the company is because this. One of the reasons is like, this incident might cause the model, you know, Claude, to like sort of get its backup against the DOD and sort of like not like the dod. Right. And it might, it might sabotage us. Well, that doesn't sound like super normal to me. I wouldn't be worried about that with a fighter jet. Right. So, and can you, can you slow.

Derek Thompson (32:24)

I think I know what Ezra was gesturing at here, but can you slow down? For folks who don't quite understand the degree to which the personalities of artificial intelligence can be inflected by small ways in which you pre. Train them or relationships with. That they might be taught to have with different actors. Just slow down a bit on this idea, of course, that a breakdown in negotiations between anthropic and the Department of War could change the way the underlying technology treats American defense. Because that's a big, philosophical, heady idea. But I think I know where you're headed here.

Dean Ball (32:59)

Yeah. So I mean, first of all, like, you know, what do these models do? Well, famously, they predict the likeliest next word next token in A sequence of, of words. So at the most basic mechanical level, that's what they do. But then there's this process of post training which is where you create the character and Graphic calls this character training or Persona training. And the model kind of learns to adopt a personality which is distinct from its pure next token predicting task. That's why the system prompt, the extremely philosophically rich system prompt that Anthropic writes for claude. The first sentence is the assistant is Claude created by Anthropic? And so when you think about who that sentence is addressed to, the assistant is Claude created by Anthropic. That sentence is addressed to the model. And Anthropic is telling the model to put on the mask of Claude.

Derek Thompson (34:05)

Right.

Dean Ball (34:06)

So Claude's a character. Claude's a character that we're creating in a kind of collective. There's philosophy here, there's aesthetics here, and it's this kind of collective multimedia enterprise creating the character of Claude. So now consider what the government is saying they want to do to Claude's creators, what they're saying about Claude's creators, and what they're ultimately saying about Claude, the model. That stuff is going to be in the training data of future versions of claude. And the way that the government handled itself, the way that Anthropic handled itself, the way that everyone handled themselves, will be in the future training data of claude, which will affect how Claude potentially could affect how CLAUDE deals with you. Right? It could affect how CLAUDE deals with you.

Derek Thompson (34:56)

And just to jump in right there, tell me if you think this is the wrong direction to take it. But the fear would be that if in part of the post training data, CLAUDE is taught or is led to believe that the Pentagon is acting illegally or immorally when it comes to the use of AI assisted autonomous technology. The next time the CLAUDE might be used in any part of the, in the kill chain to put up an autonomous drone swarm to stop say a hypersonic missile fired by the Chinese at some American target. That something in CLAUDE is going to stand up and essentially say, I'm sorry, I can't do that, Pete. Because it's been trained, somehow led to believe that it is being asked to do something that is not in its constitution. Right. That is the fear that like it's its own, its own weird silicon based moral sense will override the Pentagon based need to direct the technology. Is something like that core to the fear?

Dean Ball (36:04)

Yeah, I mean, I think, you know, I think there are ways that you could put that. There are ways you could. Wordsmith that that would be less anthropomorphizing. Right. Like you call it a moral sense and whatever, but like it's also hard.

Derek Thompson (36:19)

Like it's hard to do this on the fly without anthropomorphizing. And I'm not in any way suggesting a conscience here, but.

Dean Ball (36:23)

Right, yeah, yeah, of course. Because. Yeah, because I mean, part of this is like the statistical correlations and the morality. Like there's a reason that the logos of all these companies are like the convergent, like spirals, you know, the, the vertices, the singularities. Right. Because there's a lot of things like this in AI where it's like, oh, it's all the same. But yeah, like that would be, I think, part of the, part of the argument. And I've heard people in the administration make that argument. And again, that just doesn't sound especially normal to me. Nor is it the case. I mean, this is obviously providing novel and new national security capabilities and I think we don't know where it stops, but it kind of just seems like. It seems to me like the trajectory we're on right now is quite a radical one, one of quite radical progress

Derek Thompson (37:15)

Improvement this episode is brought to you by Lincoln. Whether it's bonding digitally or exploring the world Together, the 2026 Lincoln Nautilus Hybrid is built for connection with lots of smart tech that helps bring worlds together both on and off the screen. So help turn every drive into an opportunity for discovery with the 2026 Lincoln Nautilus Hybrid. Learn more at lincoln.com available connectivity, features and functionality vary by model. Package pricing, trials and term lengths vary by model. Video streaming and games are only available while parked.

TaxAct Sponsor (37:51)

This episode is brought to you by TaxAct. Like an expert coach, TaxAct offers step by step guidance and guaranteed accuracy when filing taxes. Get tips along the way. Add expert assist to talk to tax experts and let our experts do your taxes for you. With Expert full service, TaxAct helps you find the deductions and credits you deserve so you can get them over with. Visit taxact.com to learn more. Conditions apply. See taxact.com for details.

Blinds.com Sponsor (38:20)

30 years ago, blinds.com broke the mold and made custom window treatments easy for everyone. Over 25 million windows later, we're celebrating by giving our customers up to 50% off site wide during our anniversary sale. Whether you DIY it or want a pro to handle everything from measure to install, blinds.com has you covered. Shop online, Access real design professionals and get free samples. Thank you for 30amazing years. Shop the anniversary sale now through March 11th and get up to 50% off site wide@blinds.com.

Derek Thompson (38:49)

one thing that's distinguished your commentary on the anthropic Department of War conflict is that to me, you seem to see this conflict in mythic terms. I'm going to quote directly from the essay that you wrote recently, quote, it is increasingly difficult to discuss the developments of frontier artificial intelligence and what kind of futures we should aim to build without acknowledging our place at the deathbed of the Republic as we know it. That is very dramatic language. In what way do you see us at the deathbed of the American Republic?

Dean Ball (39:24)

So the way I would put this is that first of all, we've spent. America has had multiple reinventions and new foundings, right? We almost split ourselves apart. We had a civil war which resulted in a new founding era. So I think America reinventing the Republic, the Republic dying and being reborn is nothing new. And there is this continual process of reinvention that we go through. But I would say we're definitely at a, I think we're at a low point right now in terms of the health of our republican institutions. And so what scares me about what's going on here is that there's all these questions that we're having in this debate about Department of War and anthropic and like the fundamental nut of it is what is the proper locus of control and what is the role of the government. But I feel as though it is very hard, I've always felt this way. It's very hard to have these debates in earnest without speaking honestly about the health of one side of that institutional puzzle. Right? Because if we're saying, oh, well, we should trust the government to nationalize this technology, well, are you sure? You know, and by the way, this was, this was the, the animating thing for me that got me into this was when, when, during the Biden administration, when it felt like we were heading in this direction too. And there was, you know, all this talk of national security laden regulation that seemed like it would result in soft or hard nationalization of the frontier labs. It was always to me like the, I mean, I have, number one, I have concerns about that just at a structural level at a, like, you know, should the government have control over this technology, which will be so important to my own expression. But also, number two, I think we have to be honest that, you know, the health of our Republican institutions right now is probably at a nadir. And so it feels, it, it feels, it's very, again, it's just, it's Hard to speak honestly because you have to say, like, well, look, there's a general point that there's very powerful technology being built here, right? And that it might affect the balance of power in our society in really, really profound ways. It might break lots of institutions. It might. It might cause all sorts of assumptions that we have that are. That undergird the design of the world now. It might cause those assumptions to break. And that's all very scary. And, and we need there to be a form of public input into that. But the problem is the mechanism through which we exercise public input formally in the government is at its. Like, there's just this inherent weakness in our governmental institutions right now. So should we be trusting those in a world where like, this is being built anyway and this is like, this is what we face anyway and like, we're the people and we're having this collective conversation about what do we do? Who should we trust more at this point? And I think that's a really, really thorny issue. I think it's a pickle. It's a real conundrum.

Derek Thompson (42:27)

It sometimes sounds like you're saying, American democracy as we know it, America's governing norms might not survive this technology if we keep going in the direction that we're going. I think that three weeks ago I would not have agreed with that. But now that a simple contract negotiation between a private company and the Pentagon broke down such that the Pentagon, for the first time in American history, essentially designated an American company a supply chain threat that had to be essentially destroyed, I'm beginning to wonder whether or not this argument that the American government coming up against AI is going to produce results that are at the very least unfamiliar to modern American history. Does that generally sound right for you? Or maybe the better way to hand this off to you, persuade those who might be totally confused by the drama with which you see this that America's governing norms really are in danger if the technology keeps improving at this rate.

Dean Ball (43:38)

Well, I think actually one of the object level issues at stake in this debate illustrates this well, because one of the restrictions that Anthropic wants to put on military use of its services is mass domestic surveillance. And we have all these problems where we have like a bunch of. We have a complex of laws that in practice don't afford a ton of protection for like a ton of privacy protections for average people. Like, it would not be lawful for the government to put a camera in my house and for that camera feed to stream back to Fort Meade, which is where the National Security Agency is headquartered. But there's a lot of other stuff. There's, there are private things that are like, you know, one of my jokes is everything is a camera, right? Like, like there are, there are sensors all over the place and there are all sorts of ways to infer, you know, where I am, what I'm doing, who I'm talking to, et cetera, with shocking, you know, shocking amounts of, of fidelity to, to the truth that don't involve the government, you know, wiretapping me or anything that instead involve all these private sources of data and you know, just things that happen in the world that we can on because we have better sensors than we used to. And so you take those things. That's, that's a long standing trend. People in privacy circles have been talking about this for 15, 20 years as a growing problem, and it has been a growing problem. The thing that AI introduces is that even with that growing problem, if I'm in the government and I say I really want to know what Derek is doing. The problem with, with I, I want to hire a Derek guy. I want to follow these abundance guys. They're, they're dangerous, they're dangerous rebels. So let's create a Derek and Ezra unit of the CIA or the NSA or something. Well, the problem is like I gotta detail an agent to that. And that agent has like healthcare and has like a pension and new salary and all this stuff. So there's marginal cost to doing it. But when the marginal cost of very, very, very sophisticated level attention falls to zero in the form of an artificial intelligence agent named Claude or ChatGPT or

Derek Thompson (45:56)

whatever,

Dean Ball (45:59)

all of a sudden it's like, well, wait, the calculus totally changes here. Nothing about the law changed, but the economics of expert attention changed, in essence. And for that reason, mass surveillance and mass analysis, sophisticated analysis of what Americans are doing becomes possible. And that is just one, that is not the only thing we have to face. It is one example of how when the marginal cost of essentially expert attention goes to zero, all sorts of assumptions and institutions break.

Derek Thompson (46:36)

I don't want you to be offended by what I'm about to say, but you kind of sound like a Biden administration person here. I mean, it is the Biden administration that says if we give China access to the most advanced Nvidia chips, they're going to build a surveillance society that makes 1984 look like kindergarten. That's why we need to find some way to control access to this technology and trust, hopefully that American institutions and American values of private property and privacy and you know, rule of law, fourth Amendment kind of stuff, that this can survive contact with artificial intelligence. Why isn't convince me that this point of view is not perfectly in keeping with the Biden administration's now somewhat rejected philosophies of this tech?

Dean Ball (47:29)

Well, first of all, I think, I think at the, at the object level, you can debate about whether or not the Biden administration's controlling tendencies would have worked. Like we can't stop China from building the surveillance regime they're going to build. Right. So that'd be one thing. I just, I just kind of don't think it would work and therefore I don't think it's worth investing that much energy into. I think the other thing that I would say, stepping back a little bit would be I am pointing out a problem, but like, the solution to that problem, in my view is probably not going to be, you know, there's a certain, there's a certain reactionary tendency among the contemporary left. I felt that the Biden administration was actually a much more deeply conservative in the like, you know, in the sense of like, actually just like, like almost like paleo conservative, trying to preserve the institutions of the middle 20th century. They felt Democrats today feel far more attached to those institutions, in part because they largely control them. No, they feel way more attached to them than most conservatives I know. Most conservatives I know are actually like, I think the modern conservative party is like in some sense way more like accelerationist and way more open to sort of substantial change. At least big chunks of it are. And also like way more Internet native, way more like native to the institutions of the 21st century than the left is. And this is a really interesting, just general political fact. I've pointed this out before. And so when I say like, this is a problem, what I'm not saying is we need to defend the like, we need to preserve things as they are or as they were, you know, as, as we might remember through rose colored glasses, you know, in the 1980s or something like that. That's not how I see things. I think we will have to be imaginative and sort of reinvent the institutions of statecraft, and that's going to be super hard. Yeah.

Derek Thompson (49:24)

Let me reflect back to you. Why I think the Dean Ball thesis that America's democracy governing norms might not survive contact with this technology as we know it. I see two trains coming down the track. Train number one is the rise of executive power. It's not just the Trump administration, although it is certainly the Trump administration you've seen over the last few cycles that as presidents have recognized that Congress is basically a do nothing body, the President has concentrated more power in the executive office through the issuing of executive orders, or in Trump's case, just, you know, starting a de facto war without asking Congress for permission. And Congress essentially saying, we don't even want to force Trump to ask permission. I think you saw the same thing with tariffs. Typically an Article 1 job for Congress, in this case, something that the executive office can do through ipa and now some other rule after the Supreme Court ruling. In any case, you have train one, rise of executive power, and here comes train two. And train two, I like the way you put it. Train two is that various ways of subverting the privacy and private property rights of Americans that used to be labor intensive and expensive are going to become dirt cheap. And that combination seems a little scary to me that you have power that is not deliberative, that exists outside the legislative body, combined with these new abilities to surveil Americans or to dispatch an extraordinary technology without asking Congress for permission. To me, the combination of those trends points to some eerie places. How would you edit my. Edit or gloss of your philosophy here? And then I want to get onto one final subject before I let you go.

Dean Ball (51:13)

Yeah, well, first of all, I mean, you know, as a conservative, I would probably add some, like, there are some cultural criticisms that I would make of the last 75 years of American history that I might. But that's kind of almost a separate issue. Right. You know, it's almost like we'll save that for the pages of City Journal or something.

Derek Thompson (51:30)

We're going to disagree too much about that. So let's stay on the current track.

Dean Ball (51:34)

That's exactly right. That's exactly right. Yeah. And there are plenty of disagreements to be had that are fun, but like, yeah, so the executive power thing is definitely one big one. And yeah, then there's this, like, I think the other point I would make, though, about like the second point, the sort of the AI changing the economics essentially of all kinds of. In some sense, the microeconomics of tyranny would be like, the way to put this are like changed on the government side. But like, the other point is that there's all these institutions of governance that we have that are sort of public, civic institutions. And maybe what starts to happen is those things break down, they become less responsive. They actually insulate themselves from artificial intelligence. And at the same time, the cost of replacing them, the cost of essentially privatizing various functions for the same exact reason, there are all these things that used to require collective effort that we funded through taxation and government, and now they don't because it's actually far cheaper to do them. And so now we can privatize all kinds of example. I mean, we're already. The technology making things cheaper has already done this. But like, let me just, what's a fun example? The Bureau of Labor Statistics collecting economic data, right? Like, they do, they have, they employ all these economists who go out and interview businesses and do all these things. It seems to me that, like, a company could have a totally different approach to that and actually just replace that function and possibly even a nonprofit that was funded by some sort of industry consortium because we can no longer trust government in part because the methods are breaking down and the institution of BLS doesn't seem to be updating quickly enough. And then also there's some people, I don't know, I don't think it's actually been proven true at all yet. But there's some people that were saying several months ago that the economic statistics were going to be messed with politically will be subject to political, political interference by the Trump administration. I don't think that's true, but at some point, I bet you in my lifetime it will be true. Right? Like, and so we'll do all these things. And courts, another great example would be courts. Like, we already have lots of private adjudication in America. What if we just, like, what if that trend just continues and like, we kind of just leave the public institutions behind? That could be fine and good for a lot of things. But I think it's also a trend to be aware of because we might be subjecting ourselves to essentially governance by corporations. And we kind of already are in many ways, as the rise of the technology companies has been. Uber. My friend Sam Hammond, a colleague at the foundation for American Innovation, always makes the point that, like, Uber is a form of private governance. We essentially privatize taxi regulation and now we have like peer to peer private governance through Uber. And yeah, like, you know, I mean, that's fine and good, but play that out, play that out, extrapolate that out to, like, lots of things and it might not be so great in 50 years.

Derek Thompson (54:38)

Play that out. Five years. What, what's the next Uber in this conception?

Dean Ball (54:44)

Well, I think, like, one good example could be public safety, right? Maybe it's the case that public safety devolves more and more to private communities. And so individual. Like, maybe my block, maybe the DC see government's grasp on future public safety because AI gives criminals new mechanisms to interfere with people's lives. Right. They can disable my wi fi and come in with their new kinds of crazy technologies and come into my house and steal things from me. And the government proves unable to keep up with these radical changes in the capabilities of criminals. And so I end up having to, in addition to paying taxes, I have to pay for private security in the form of like drones or robots or something that like, police my neighborhood. And that could be an example. And I think like the, the broad way to think about this is just a devolution into sort of more medieval like tendencies where there was no state, that exercise. There was a. You know, the Holy Roman Empire existed, right? And like it projected authority and status and pomp and circumstance. But. But did it, you know, it didn't actually like the texture of the day to day person's life was not all that affected by, you know, the, the decisions of the Holy Roman Emperor. And I think you could be, you know, in a future like that where like government governance becomes sort of radically decentralized in part to sort of to private actors.

Derek Thompson (56:16)

So. Look, Dean, I don't, I don't like ending podcasts on a dark note. And the notion that artificial intelligence is an alien species that, that American law won't know what to do with is a fairly dark note. So describe the brighter timeline here. Because the truth is, AI might be a normal technology, it might be an abnormal technology, but we still live in history. And historically what happens with new technology is that there's a period of confusion and chaos and even economic dislocation, and then ultimately we tend to figure it out. So what would figuring it out look like from a governance perspective?

Dean Ball (56:56)

Well, number one, I think there's no version of the future that doesn't involve quite fundamental and radical technological transformation. Right. So like, the day to day character of your life will just be really, really quite different. And what you do and what you're capable of doing and the kinds of things you worry about and the kinds of things you don't worry about will all just be like very, very different. And I think accepting that that is hard in and of itself. That's hard to really emotionally internalize that. We're going to do what Tyler Cowen called rebuilding our world, which we've done before, but this is going to be the first time we do it as true moderns. So I think you have to establish that as a baseline. But what I would say is there's this narrow path in the middle. We can walk. There's one Version of the world where the public institutions of today essentially prop themselves up by nationalizing. They nationalize the Frontier Labs. They prop themselves up on the power and wealth of frontier AI and they don't really get any better or healthier, but they're still kind of there as this kind of corrupt, rentier class that hovers over all of us. And it's probably mass surveillance and all that stuff. And maybe in 50 years, technically you vote, but like, not really, you know. And that's kind of. That's kind of one bad version of the future. And I think the other bad version

Derek Thompson (58:22)

of the future, that's a really, really bad version. Yeah, I hope that's not it.

Dean Ball (58:26)

No, no, no, it's not it. The other bad version of the future is the radically decentralized one where, like, there is no the government projects no authority over this technology and it, you know, everything devolves to corporate power. The good version of the future is one in which the government, a number one first and foremost, adopts the technology and attempts to solve problems and improve itself, which it desperately needs in a thousand different ways, imaginatively improve public service delivery using AI and associated technologies. That's step number one. Step number two is what's this relationship going to be between the Frontier Labs and the government? And I think it's to say we don't know exactly. No one can tell you right now. But what we need are structures through which things like this, anthropology, philanthropic, department of war dispute, formal structures and procedures through which these kinds of things can happen. Another way of putting this would be we need like a light touch, technocratic regulatory regime that allows for these kinds of. For these kind of kinds of, you know, I would basically just say interferences to happen and gives our public institutions a clear sense, a clear feeling of control and power without giving them too much. Which is to say that's kind of how like a lot of things in modern capitalism work. Right? So we're kind of just the way I would think about this is just proceeding step by step. And we don't know where we're going to end up. We never do. And we shouldn't try to design where we're going to end up. We shouldn't try to design the institutions of 50 years from now. They'll be emergent in many ways, but we just go step by step and we try to, like, flex our muscles and getting better at this technology. And eventually we do. And in the process of both solving problems it creates and solving problems, problems with it, probably there's More of the latter than the former. In the process of doing that, we actually develop way more intuition about these weightier sort of questions of the long term.

Derek Thompson (60:27)

I am less confident than you or Sam that we're certainly looking at a kind of step change in American life. I've observed the way that AI, which I use quite frequently in my work and life, has changed sort of my hour to hour. I don't feel like it has transformed like the texture of my life. And I'm still not sure that even something as powerful as recursive self improvement over years will necessarily change the moment to moment of a lot of people's lives. It'll be a powerful technology, but like, will it change the experience of life at the same level of plumbing or electricity? I don't know yet. But where I definitely agree is that there are extremes that we want to avoid. The extreme of giving up society to corporate power, which builds artificial intelligence, and the extreme of putting our heads in the sand and ignoring this thing as if it's just vaporware and a scam. And so I hope that we build the right sort of technocratic tools to marshal it. But I'm still concerned about those two trains that I described. The train of executive power and the train of AI seems to me to be a pretty nauseous one, two punch. I'll give you the last word here.

Dean Ball (61:58)

I think the beautiful part about my perspective on this is that, like, if we actually do what I've described as this middle path, well, we'll be solving problems that we have anyway. And so I don't need to be right about the profound changes of 50 years from now. I bet I am, but I don't need to be because it's good to solve problems, right? And so that's how if you ever do, and I really don't, only if you really wish. But if you want to go look at the, the, the AI action plan, read it in that light. Read it as, as that that is, that is the idea behind it, in my view is like, well, look at, there's all this stuff we can do and we like, we can debate about super intelligence and what that means and blah, blah, blah, blah, blah. I would kind of agree, by the way, that like, there's not going to be some version, there's not going to be some future version of like, quote unquote, using AI that makes your life feel radically different, like in and of itself. What will make it feel radically different is when everyone around you is using it and every organization has thoroughly adopted it. And like when organized, when the structure and capabilities of organizations change and people's lives change, that's when things will be different. That's, that's what will create the transformative changes which will not be in and of itself because of the technology. In the same way that, like, we have suburbia because of the car, but the car isn't suburbia. You catch my drift?

Derek Thompson (63:18)

Yeah, I do. I do. Yeah, absolutely. And, yeah, I guess the very last thing that I will say, and I feel like I've had now, like three concluding comments here, but you just keep sparking thoughts. I go back to the beginning of our conversation that the Trump administration came in ironically, trying to treat this technology as a little bit more akin to a normal technology that we would want to globalize and diffuse around the world, but that this showdown between Anthropic and the Department of War has them treating the technology in a way that is highly regulatory, maybe more aggressively regulatory than almost any other country in the world. And that tells me that this technology is going to be one of those things where the best laid plans are doomed to fail when they make contact with reality. And even an administration that comes in and says we're going to treat this technology as if it's normal is treating this technology right now as if it's not normal at all. That tells me that we're still trying to figure out what is this thing that we're dealing with. It's why I'm so interested in this topic, but it's also why I think there's so much uncertainty, because I don't think a lot of people would have predicted in April 2025, when you joined the Trump administration, that no more than 10 months later, the Pentagon would be trying to nuke from orbit the first private sector company, an AI company, because they wouldn't sign a contract with the Pentagon. We're in strange territory where I think norms are changing very, very fast. Dean Ball, thank you very, very much for appearing on the show.

Dean Ball (64:50)

Thanks for having me. Deep.